Symbioza

by dev.symbioza

Overview Discussions Admin

Server Details

Run a GPU job under a hard dollar cap and collect the files it writes.

Ownership verified

Status: Healthy
Uptime: 99.2% over 21 days
OAuth: Works in Glama
Last Tested: 2026-10-19 21:34
Transport: Streamable HTTP · MCP 2025-11-25
URL
Repository: symbioza/claude-plugin
GitHub Stars: 0

TDQS

A4.4/5.0

Scored 2026-10-17 09:30 across 9 tools

Disambiguation4/5

getStatus and inspectRun both read a running job's progress by executionId, so an agent may initially need to choose between them; otherwise purposes are clearly distinct across estimate, submit, cancel, artifact, listing, and dataset measurement. The very detailed descriptions help resolve the overlap.

Naming Consistency5/5

All nine tools use consistent camelCase verb+noun naming (cancelJob, describeDataset, estimateExecution, getArtifact, getStatus, inspectRun, listJobs, listSecrets, submitJob). No mixed conventions or vague one-word names are present.

Tool Count5/5

Nine tools neatly cover the job lifecycle without obvious bloat or thinness. Each tool earns its place for pricing, submission, monitoring, cancellation, artifact retrieval, listing jobs/secrets, and dataset description.

Completeness5/5

The surface covers estimate -> submit -> monitor -> artifact -> cancel, plus job/secret listing and dataset measurement, with retry and dedupe handled inside submitJob. Missing secret creation is intentional because users save secrets via the web UI, not a critical lifecycle gap.

Available Tools

9 tools

cancelJobCancel a job and tear down its machineA

DestructiveIdempotent

Inspect

Cancels the job and tears down its machine. Symbioza stops your workload and collects the files it has written to /workspace/artifacts/ before the machine is released, within a short bounded window; the copies Symbioza already took during the run are kept even if that last collection cannot finish, getArtifact returns them marked recovered, and a file that could not be collected is listed as not delivered. If we confirm the stop, you are charged for measured time up to then, and never for more than a minute past the moment you asked; if the machine cannot be reached, you are charged only up to the moment you asked. Never more than your spending limit. getStatus shows the request as cancelRequestedAtMs at once (cancelState requested), and the job reads cancelled only once the stop is CONFIRMED (cancelEnforcedAtMs set, cancelState enforced); if the stop cannot be proven the job reads lost with cancelState unconfirmed and the last evidence in cancelStopDetail. Pass an optional reason (short human-readable text, truncated at 500 chars) to record WHY; it is stored opaquely and returned by getStatus as cancelReason so a client can see why a job was cancelled.

ParametersJSON Schema

Name	Required	Description
`reason`	No	Optional human-readable reason for the cancel, surfaced to the client via getStatus.cancelReason. Truncated at 500 chars; stored opaquely, never parsed.
`authKey`	No	Usually omit this. The customer door — the OAuth-signed-in HTTPS connector at https://symbioza.dev/mcp — authenticates the session itself, so no key is passed per call. Only a self-hosted or test transport passes a key here.
`executionId`	Yes

Output Schema

ParametersJSON Schema

Name	Required	Description
`note`	No
`reason`	No
`status`	No
`cancelled`	Yes
`executionId`	Yes
`cancelReason`	No
`deadlineAtMs`	No
`alreadyTerminal`	No
`workloadFinished`	No

TDQS

A4.5/5.0

Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Goes well beyond the destructive/idempotent annotations by disclosing the cancellation state machine (cancelRequestedAtMs -> cancelEnforcedAtMs -> cancelled/lost), artifact collection guarantees, and precise billing semantics. This is exactly the extra context annotations cannot carry.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness3/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The core action is front-loaded, which is good, but the remainder is a dense, unbroken wall of text mixing state transitions, billing, and artifact rules. The content is valuable, yet a list or sentence breaks would make it scannable without losing information.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a destructive, open-world job-cancellation tool with an output schema, the description covers the lifecycle, timing, billing, and reason handling an agent needs before invoking it. Nothing material is missing.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 67% and already documents reason and authKey, but the description reinforces reason's meaning by tying it to getStatus.cancelReason and its 500-char opaque storage. executionId remains undocumented in both places, keeping this just under a 5.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

States a specific verb and resource ('Cancels the job and tears down its machine') and names the sibling tools it interlocks with (getStatus, getArtifact). An agent can distinguish this from submitJob or getStatus without opening a schema.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

It clearly frames when this tool is the right action and points to getStatus and getArtifact as the ways to observe the outcome. It stops short of an explicit when-not or cost/benefit comparison against alternatives, so it is strong context rather than full routing guidance.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

describeDatasetMeasure a dataset URL (free, books nothing)A

Read-onlyIdempotent

Inspect

Measure a public https dataset URL and return its exact sizeBytes, sha256 and contentType, plus suggestedEntry — a filled-in element of the spec's datasets[] array you can paste straight in. Call it before submitJob whenever you want datasets[] to stage and verify a file: sizeBytes and sha256 are REQUIRED on every entry, and this is how you get them without pulling the data down to your own machine first. We read the URL from our side, stream it through a hash and keep no copy. Free, books nothing, and it never reads the caller's filesystem. https only, and only addresses on the public internet: a loopback, private, link-local or otherwise reserved host is refused, and so is a redirect that leads to one. There is an upper size limit and a limit on how often one session may call it; both refusals say so plainly. You do NOT need this when your command downloads its own data with curl or wget — datasets[] is the verified-staging path, and only that path needs a hash declared up front.

ParametersJSON Schema

Name	Required	Description	Default
`url`	Yes	Public https URL of the dataset file. Symbioza fetches it from its own side — your own machine is never on the data path, and a URL that needs a credential to read should be a signed URL with the shortest expiry that covers staging — once submitted it is stored with the job spec in clear and included in our backups.
`authKey`	No	Usually omit this. The customer door — the OAuth-signed-in HTTPS connector at https://symbioza.dev/mcp — authenticates the session itself, so no key is passed per call. Only a self-hosted or test transport passes a key here.

Output Schema

ParametersJSON Schema

Name	Required	Description
`code`	No	Machine-readable refusal code: unauthorized, invalid_spec, budget_too_low, insufficient_balance, access_required, secret_conflict, no_compute, unavailable, refused
`action`	No	The one next step: authenticate, fix_spec, raise_budget, add_funds, claim_credit, add_access, relax_spec_or_retry, retry_later, contact_support
`reason`	No	Why, in a customer-safe sentence
`sha256`	No
`rejected`	No	Present and true when the call was refused; nothing was booked or charged
`sizeBytes`	No
`contentType`	No
`suggestedEntry`	No	A datasets[] element for the spec, ready to paste
`supportedImages`	No	On an unverified-image refusal: the images the door accepts

TDQS

A4.8/5.0

Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations already declare readOnly/openWorld/idempotent/non-destructive, yet the description adds substantial context beyond them: server-side streaming with no retained copy, no caller filesystem access, https-only with loopback/private/link-local/reserved hosts and redirects refused, plus size and rate limits whose refusals are self-explanatory. This is genuinely useful operational detail.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

It is long but front-loaded with the core purpose and returns, and nearly every sentence carries a distinct rule (safety, limits, exclusion). The density is high, though a couple of clauses (backups/clear-storage asides) could be trimmed.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Output schema exists so return values needn't be detailed, yet they are summarized; the invocation preconditions, safety profile, and the alternate non-hashing path are all covered. Nothing needed to call it correctly is missing.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 100% and the schema descriptions are already rich (signed-URL guidance, when to omit authKey). The description still adds meaning about the url parameter's constraints — public-internet-only, https-only, refusal semantics — which is not in the schema itself.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The title and first sentence state a specific verb (measure) and resource (a public https dataset URL) and enumerate the exact returns (sizeBytes, sha256, contentType, suggestedEntry). It is immediately distinguishable from siblings like submitJob and estimateExecution.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

It names the trigger condition (call before submitJob whenever datasets[] must stage and verify a file) and an explicit exclusion ('You do NOT need this when your command downloads its own data with curl or wget'). That is textbook when/when-not guidance.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

estimateExecutionEstimate a job's price (free, books nothing)A

Read-onlyIdempotent

Inspect

Symbioza price, cost, feasibility, balance or credit: ALWAYS call the tool (estimateExecution for a price), never answer from memory, prior chat, arithmetic or cached values, never replace its price card with prose, never invent a total. Never make the user pick a budget just to see the likely cost: estimate without budgetUsd, explain the estimate, then ask for the spending approval. Call estimateExecution once per spec and budget: the answer already has the price, runtime, readiness and balance; call again only when the spec or budget changes. Symbioza selects compatible compute internally and returns estimate, shown first: status best_effort with likelyUsd, lowUsd–highUsd, confidence, runtime (the expected runtime) and costBreakdown, or rate_only with only perHourUsd; then price (evidence, summary, perHourUsd, includes, unknown, nextStep), estimatedPriceUsd (only when price.evidence is customer, then equal to estimate.likelyUsd), readiness (preliminary while a budget or saved access is missing — budget_required, access_required — else ready; inputs_unverified and inputs_not_declared are advice that never blocks a submit) and runtime (boundSeconds, boundBy, note): only where the run would be stopped — a limit, never a prediction. A maxRuntimeSeconds above 172800 (48 hours) is refused here too, with invalid_spec, as submitJob refuses it. No matching machine: a refusal in submitJob's shape — reason, code no_compute, action relax_spec_or_retry; still free and still books nothing. image must be one of the images we have verified launch on our machines: pytorch/pytorch:2.4.1-cuda12.1-cudnn9-runtime, pytorch/pytorch:2.4.0-cuda12.1-cudnn9-runtime, tensorflow/tensorflow:2.15.0. Any other image is refused with code invalid_spec, action fix_spec and a supportedImages list, before anything is booked or charged. budgetUsd is what you are willing to spend on the job, and you are never billed more than that. Every dollar figure returned by any tool here is in that same unit: what you pay. It also returns account: availableUsd (what this account can spend right now, promotional credit included), promoUsd and promoExpiresAtMs, promoJobCapUsd (the most promotional credit any one job may use — a larger budget needs paid balance for the rest), coversThisJob (would submitJob accept this spec's budget today; null when the spec has no budgetUsd), claimableUsd (free compute credit this account could still take, or null), nextAction ("add_funds" or "claim_credit", null when the job is already covered), claimUrl, and with a budget topupUrl and shortfallUsd (the whole dollars of credit this job still needs, 0 when covered; topupUrl carries it — give the user that link as returned). When the spec attaches secrets, it also returns access: ready (the saved sets that will be injected: name, type and variable names), required (each missing, ambiguous or type_mismatch reference — a missing one with a setupUrl to give the user, an ambiguous one with candidates to ask about), conflicts, complete and nextAction (add_access or fix_spec); submitJob refuses until access.complete is true. A warning env_may_hold_credential names env variables that look like credentials — move them to saved Secrets. It also returns specDigest, a fingerprint of the exact spec and budget priced here — only when the spec carries budgetUsd, since submitJob refuses one without it: pass it to submitJob and a spec or budget changed since this estimate is refused with invalid_spec instead of run. estimate.runtime is the one runtime story: how long this job likely runs (likelySeconds, its range and one evidence sentence); runtime.boundSeconds is only where the run would be stopped — a limit, never a prediction. When estimate.status is best_effort, show estimate.likelyUsd with its range (lowUsd–highUsd), the expected runtime, confidence, costBreakdown and disclaimer — a best-effort estimate from the strongest evidence about this workload; it is not a quote, and image history alone is never evidence about this job. When estimate.status is rate_only, nothing predicts the runtime: tell the user there is no total, and give price.perHourUsd (what each hour of running costs). estimatedPriceUsd is returned only when price.evidence is customer — the estimate priced at the user's own maxRuntimeSeconds — and then equals estimate.likelyUsd. No range is given without evidence for one. budgetUsd stays the hard maximum either way, and is never needed to see an estimate. Read the account before submitJob so a funding problem is something you fix rather than something you discover from a refusal — claiming credit needs no payment. Declare datasetGb as the total data the box will ingest, including data it downloads itself, so a host too slow to pull it in time is filtered before booking. Declare gpu.minCudaVersion (e.g. 13.0) if your image pins a CUDA build — it filters the market here too, so the price you get back is for a host that can actually run your image. Only when the user explicitly requires a particular model for their own benchmark, declare gpu.gpuNameExact: true with gpu.gpuName to price only that exact card model, with every substitute refused — the price then reflects the instrument a measurement actually needs. Add gpu.minPowerLimitW to price only boards at or above that TDP, since one model name covers several. May also return advisory warnings/notes (additive, may be absent) — e.g. ingest_may_exceed_deadline when the declared datasetGb cannot move inside maxRuntimeSeconds over the link the selected host advertises (or the rate we measured on that box before), and ingest_not_predicted when that host publishes no link speed and we have never measured one. Advisory ONLY: it never refuses and never changes the ranking or price; it is about the box we would book for you, not about your data's source, whose speed we do not measure from here.

ParametersJSON Schema

Name	Required	Description	Default
`spec`	Yes	The job spec to estimate: image + command run on a rented box sized by minDiskGb (and gpu, when declared). budgetUsd is optional here — leave it out to see the likely cost before the user chooses one; submitJob requires it. Supply the workload: the image, the command, its inputs, the expected outputs and budgetUsd. Symbioza sizes the machine — do not ask the user to choose CPU cores, RAM, VRAM, a GPU model, a provider or a region, and do not derive a resource minimum from a thread count, a worker count or a measurement taken somewhere else. gpu.minVramGb, minRamGb and minCpuCores are optional expert floors: declare one only when the workload has evidence for it, and omit it otherwise — an omitted floor is sized by Symbioza, never treated as zero. The gpu block is optional: leave it out and the job is still GPU work with Symbioza sizing the card; declare it to carry minCudaVersion or an evidenced floor.
`authKey`	No	Usually omit this. The customer door — the OAuth-signed-in HTTPS connector at https://symbioza.dev/mcp — authenticates the session itself, so no key is passed per call. Only a self-hosted or test transport passes a key here.

Output Schema

ParametersJSON Schema

Name	Required	Description
`code`	No	Machine-readable refusal code: unauthorized, invalid_spec, budget_too_low, insufficient_balance, access_required, secret_conflict, no_compute, unavailable, refused
`notes`	No	Advisory only, never a refusal
`price`	No	What the price rests on, what it includes and what is still unknown
`access`	No	The saved secrets this spec attaches, by name and type only
`action`	No	The one next step: authenticate, fix_spec, raise_budget, add_funds, claim_credit, add_access, relax_spec_or_retry, retry_later, contact_support
`reason`	No	Why, in a customer-safe sentence
`account`	No	This account's spending position for the spec
`runtime`	No	boundSeconds, boundBy (budget \| customer \| platform), note: where the run would be STOPPED on that machine — a limit, never a prediction. The expected runtime is estimate.runtime
`estimate`	No	The best-effort estimate of the total bill: show likelyUsd with its range, confidence and disclaimer, or perHourUsd when rate_only. budgetUsd stays the hard cap
`rejected`	No	Present and true when the call was refused; nothing was booked or charged
`warnings`	No	Advisory only, never a refusal
`budgetUsd`	No	Your own cap, read back; nothing bills past it. Absent when the spec has none
`readiness`	No	What stands between this estimate and a submittable one, plus advice about its inputs
`provenance`	No
`specDigest`	No	Fingerprint of the exact spec and budget priced here; pass it to submitJob. Absent when the spec has no budgetUsd: submitJob requires one
`supportedImages`	No	On an unverified-image refusal: the images the door accepts
`estimatedPriceUsd`	No	What this spec is expected to bill you, in dollars — present only when price.evidence is customer (the estimate priced at your own maxRuntimeSeconds), and then equal to estimate.likelyUsd; otherwise absent and the total is estimate.likelyUsd

TDQS

A4.7/5.0

Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations already declare readOnly/idempotent/non-destructive, but the description goes far beyond them: it discloses that the call is free and books nothing, that maxRuntimeSeconds above 172800 is refused with invalid_spec, the no_compute and invalid_spec refusal shapes, the exact verified-image allowlist, and that specDigest gates submitJob. These are behavioral facts no annotation or schema conveys.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness3/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The critical rule ('ALWAYS call the tool, never answer from memory') is front-loaded, which is good, but the body is a very long wall of text with visible redundancy — the runtime-boundary sentence ('only where the run would be stopped — a limit, never a prediction') and the budgetUsd hard-maximum point each appear more than once. Several sentences do not earn their place.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a tool this complex — nested spec, secrets, account/access blocks, advisory warnings, specDigest chaining into submitJob — the description leaves nothing an agent needs to call it correctly. It even explains return values despite an output schema existing, and covers the refusal paths and funding/access recovery actions.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

With 100% schema coverage the baseline is 3, but the description adds real meaning: budgetUsd is 'what you are willing to spend' and never exceeded, datasetGb must include data the box downloads itself, and gpu.minCudaVersion/gpuNameExact/minPowerLimitW are explained in terms of what market filtering they cause. It stops short of documenting every parameter, but the schema already covers the rest.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description states a specific verb and resource (estimate a job's price) and adds the decisive scope fact — it is free and books nothing — which separates it from submitJob. The opening line even names the tool and its output ('estimateExecution for a price'), so an agent can distinguish it from getStatus/listJobs without opening a schema.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

It gives explicit when/when-not rules: always call this rather than answering from memory, call once per spec+budget and only re-call when either changes, estimate first without budgetUsd and then ask for approval. It also routes to siblings by name, telling the agent to read the account before submitJob and to use listSecrets for set names.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

getArtifactCollect a finished job's resultsA

Read-onlyIdempotent

Inspect

Results of a finished job: your workload's exitCode, the artifact manifest (path/sha256/status, with time-boxed download links when offsite storage is configured), deliveryComplete + deliveryDetail, priceUsd (the charge, the same number getStatus returns) with priceStatus ("provisional" while the provider's bill is unconfirmed, "final" once reconciled or when nothing was charged; never above budgetUsd), checkpoint. Each download link is for the user's browser, where a video, image, PDF or text file opens directly and anything else downloads. Your own environment may not be able to fetch it: to inspect a file's contents, give the user the link and ask them to attach the downloaded file to the conversation. A link you cannot reach is not a missing file: deliveryComplete and the manifest say what arrived. The charge shown after completion may change while billing is reconciled. Your total charge for the job, including retries, will not exceed your approved budget. Status and artifact reads return text only and draw no card: a job's card is the one on its submitJob result (and the price card on estimateExecution), and that card keeps itself current while the job runs.

ParametersJSON Schema

Name	Required	Description	Default
`authKey`	No	Usually omit this. The customer door — the OAuth-signed-in HTTPS connector at https://symbioza.dev/mcp — authenticates the session itself, so no key is passed per call. Only a self-hosted or test transport passes a key here.
`executionId`	Yes

Output Schema

ParametersJSON Schema

Name	Required	Description
`reason`	No
`storage`	No
`exitCode`	No
`priceUsd`	No
`artifacts`	No	The delivery manifest: files under /workspace/artifacts/
`available`	Yes
`checkpoint`	No
`executionId`	Yes
`priceStatus`	No
`transferOut`	No
`workloadLog`	No	The tail of your workload's own stdout/stderr
`deliveryDetail`	No
`executedAttempt`	No	The attempt that ran your command (0-based); null when no attempt ran it
`deliveryComplete`	No
`priceStatusReason`	No	reconciled \| awaiting provider confirmation \| nothing was charged
`recoveryAssessment`	No

TDQS

A3.9/5.0

Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Goes well beyond the annotations: time-boxed download links, links intended for the user's browser, the caveat that the agent's own environment may not be able to fetch them, that a fetch failure is not a missing file (deliveryComplete/manifest are authoritative), that the charge may change during reconciliation, that retries cannot push the total above budgetUsd, and that no card is drawn on this read.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness3/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Front-loaded correctly with the payload contents, and most sentences carry operational weight, but it is one dense paragraph that runs long and repeats the budget ceiling twice ('never above budgetUsd' then 'will not exceed your approved budget') plus restates the card behavior in two sentences.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

An output schema exists, so explaining return fields is technically redundant, yet the description's real value — how to handle browser-only download links and why the charge can move — is not derivable from the schema or annotations. Given moderate complexity and full annotation coverage, the definition is complete enough to call correctly, with only the executionId provenance left thin.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters2/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 50%: authKey is documented in the schema but executionId has no description anywhere. The description never explains the identifier's meaning, format, or where to obtain it (only indirectly referencing the submitJob result), so for a tool keyed on a single required ID it fails to compensate for the coverage gap.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

States a specific resource and verb-equivalent ('Results of a finished job') and enumerates the payload: exitCode, artifact manifest with path/sha256/status, deliveryComplete, priceUsd, checkpoint. It also distinguishes itself from the sibling getStatus by noting priceUsd is 'the same number getStatus returns', so an agent can route between the two.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines3/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Usage is implied by 'a finished job' (call after completion, not mid-run) and it gives concrete operational guidance for download links. However, it never explicitly states when to choose this over getStatus or inspectRun, nor what prerequisite state the job must be in, leaving the sibling choice to inference.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

getStatusJob statusA

Read-onlyIdempotent

Inspect

Progress of a submitted job by executionId: status/stage, units, budgetUsd, deadline, heartbeat age, transfer, storage, incidents. Status: queued = waiting for a machine (nothing booked or charged yet); running; completed = the command finished and every file was delivered (the web shows it as 'delivered'); a finished command with missing files is failed with deliveryOutcome partial — collect the files with getArtifact before considering a rerun; cancelled; lost. Status and artifact reads return text only and draw no card: a job's card is the one on its submitJob result (and the price card on estimateExecution), and that card keeps itself current while the job runs. budgetUsd is what you are willing to spend on the job, and you are never billed more than that. Once the job is over, priceUsd is what it has billed — including a run that failed or was cancelled, and never above budgetUsd — and priceStatus says whether that number can still move: "provisional" while it awaits provider confirmation, with no time bound promised; "final" once it is reconciled against the provider's own rows, or when nothing was charged. The charge shown after completion may change while billing is reconciled. Your total charge for the job, including retries, will not exceed your approved budget. artifactsReady is true once the job has finished and at least one file was delivered; getArtifact's deliveryComplete says whether every file arrived. nextAction is one machine-actionable hint from a fixed set ("poll getStatus", "getArtifact", "getArtifact; delivery incomplete, N of M files delivered", "getArtifact; delivery unconfirmed", "getArtifact; files recovered after the job stopped", "read cancelReason, fix, resubmit", "check incidents; workload failed, fix and resubmit", "check incidents; nothing of yours ran, resubmit", "check incidents; resubmit", "resubmit", "blocked: read recoveryAssessment; a new run needs a new estimate and approval"), or null. Read it before you touch the spec: a failed job says "nothing of yours ran" when every attempt died before your command started (provisioning, verification or data staging), and "fix and resubmit" only when your own workload ran and failed; either reads "blocked: read recoveryAssessment; a new run needs a new estimate and approval" instead when its recoveryAssessment calls the job blocked (a refused download, a pinned machine, a booking rule, a spending limit, an unconfirmed release). recoveryAssessment says what can happen next (mode none, resume, restart, blocked or user_stopped, with a reason and a next step in words); it never acts on its own, and any new run needs its own estimate and your approval. When status is "cancelled", cancelReason is your own reason when you cancelled (null if you gave none), else a note from Symbioza for you, else the fixed sentence "Stopped by Symbioza before it finished. You are charged only for the time used; the incidents say what was measured."; nextAction reads "read cancelReason, fix, resubmit" only when it carries your own reason or a note. priceStatusReason says which rule decided priceStatus ("reconciled", "awaiting provider confirmation" or "nothing was charged"). deliveryOutcome says what became of the files once the job is over ("delivered", "partial", "none" or "unconfirmed"), null while it is not. While queued, capacityWait.text says where the job stands: behind a real capacity gate "Waiting for a machine · 3rd of 3 in line · since 09:07 UTC · if none is free by 15:07 UTC the run ends and nothing is charged", otherwise "Queued · waiting for a machine · 2nd of 2 in line". cancelOrigin says who stopped it: "own" (you), "symbioza", or null when nothing was cancelled. Poll after submit.

ParametersJSON Schema

Name	Required	Description	Default
`authKey`	No	Usually omit this. The customer door — the OAuth-signed-in HTTPS connector at https://symbioza.dev/mcp — authenticates the session itself, so no key is passed per call. Only a self-hosted or test transport passes a key here.
`executionId`	Yes

Output Schema

ParametersJSON Schema

Name	Required	Description
`found`	Yes
`stage`	No
`reason`	No	Why nothing was found (another account's id reads the same as an unknown one)
`status`	No	queued \| running \| completed \| failed \| cancelled \| lost
`storage`	No
`attempts`	No
`complete`	No
`exitCode`	No	Your workload's own exit code
`priceUsd`	No	What the job has billed once it is over; never above budgetUsd
`recovery`	No
`budgetUsd`	No	The cap you set, in your units
`incidents`	No
`nextAction`	No
`transferIn`	No
`cancelState`	No	none \| requested \| enforced \| unconfirmed
`executionId`	Yes
`priceStatus`	No	provisional \| final
`cancelOrigin`	No	own \| symbioza — who stopped the job; null when it was not cancelled
`cancelReason`	No
`capacityWait`	No	position, of, startedAtMs, closesByMs, text — while queued for capacity
`deadlineAtMs`	No
`unitsPlanned`	No
`workerStatus`	No
`artifactStore`	No	offsite (download links) \| local
`artifactsCount`	No
`artifactsReady`	No
`deliveryDetail`	No
`heartbeatAgeMs`	No
`machinesBooked`	No	Distinct machines this job held; attempts also counts any refused before a machine was booked
`unitsCompleted`	No
`deliveryOutcome`	No	delivered \| partial \| none \| unconfirmed once the job is over; null before
`executedAttempt`	No	The attempt that ran your command (0-based); null when no attempt ran it
`cancelStopDetail`	No
`deliveryComplete`	No
`lastHeartbeatAtMs`	No
`priceStatusReason`	No	reconciled \| awaiting provider confirmation \| nothing was charged
`cancelEnforcedAtMs`	No
`machineEverExisted`	No
`recoveryAssessment`	No
`cancelRequestedAtMs`	No
`workloadEverStarted`	No

TDQS

A4.3/5.0

Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations already cover the safe-read profile (readOnlyHint, idempotentHint, non-destructive), and the description adds substantial context beyond them: text-only responses that draw no card, budget capping semantics, priceStatus provisional/final behavior, recoveryAssessment modes and its non-acting nature, and cancelReason/cancelOrigin rules. This is exactly the kind of state-transition and billing detail the agent needs and cannot get from structured fields.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness3/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The design is organized field-by-field and front-loads the core purpose, but it drifts into a wall of text with real duplication — the budget cap is stated twice ('you are never billed more than that' and 'will not exceed your approved budget') and the approve-a-new-estimate point appears twice. Trimming the repeats would improve scanability without losing meaning.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Despite an output schema existing (so return values need no explanation), the description anticipates the agent's real questions for a long-running-job status tool: what each status means, what a 'partial' failure implies, how billing can still move, and what recoveryAssessment can and cannot do. Nothing needed to poll and interpret this tool correctly is missing.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 50%: authKey is documented in the schema while executionId is not. The description only says the job is identified 'by executionId' and does not add format, sourcing, or validity details (e.g., where the id comes from, whether it can be reused after a rerun). With the schema doing partial lifting, this sits at the baseline.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The opening sentence states a specific verb and resource ('Progress of a submitted job by executionId') and immediately enumerates the returned fields (status/stage, units, budgetUsd, deadline, incidents). It also routes the agent away from itself for related work ('collect the files with getArtifact'), so it is separable from siblings like getArtifact and listJobs.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

The description gives clear operative context — 'Poll after submit', read nextAction before acting on the spec, and collect files with getArtifact before considering a rerun. It explains the conditions under which nextAction takes specific values, but never explicitly contrasts itself with listJobs or estimateExecution, so it stops short of full when/when-not routing.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

inspectRunInspect a running job (health, progress, logs, files)A

Read-onlyIdempotent

Inspect

One read of a job while it runs, by executionId. Three separate answers: connectivity (connected | silent | never | ended — whether the machine's agent is reporting), telemetry (fresh | stale | absent — whether the workload's progress reports are arriving) and progress (advancing | expected_quiet | suspected_stall | unknown). progress is advancing only when the workload's own unitsCompleted rose recently; a heartbeat alone is never progress. suspected_stall appears only when the workload set its own stallAfterSeconds, and nothing is stopped because of it. Also: phase, unitsCompleted/unitsTotal, metrics, lastProgressAtMs (when units last rose, by Symbioza's clock), checkpoint.state (none | claimed | durable | loadable | torn), resume (reconnect | restart_in_workspace | resume_from_checkpoint | restart_only | unknown), the tail of your workload's output with saved-secret values redacted, the files collected so far, and capabilities, which says for each steering action whether it is available. A workload reports progress by writing /workspace/artifacts/.symbiosa/progress.json.

ParametersJSON Schema

Name	Required	Description	Default
`authKey`	No	Usually omit this. The customer door — the OAuth-signed-in HTTPS connector at https://symbioza.dev/mcp — authenticates the session itself, so no key is passed per call. Only a self-hosted or test transport passes a key here.
`executionId`	Yes

Output Schema

ParametersJSON Schema

Name	Required	Description
`seq`	No
`unit`	No
`found`	Yes
`phase`	No
`reason`	No
`resume`	No
`status`	No
`logTail`	No
`metrics`	No
`progress`	No
`artifacts`	No
`telemetry`	No
`checkpoint`	No
`unitsTotal`	No
`designState`	No
`executionId`	Yes
`capabilities`	No
`connectivity`	No
`observedAtMs`	No
`heartbeatAgeMs`	No
`progressReason`	No
`unitsCompleted`	No
`lastProgressAtMs`	No
`stallAfterSeconds`	No
`expectedQuietSeconds`	No
`progressReceivedAtMs`	No

TDQS

A3.9/5.0

Behavior4/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations already declare readOnly, idempotent, non-destructive, and closed-world, so the safety bar is covered. The description adds rich context beyond them: that suspected_stall stops nothing, that secret values are redacted in the output tail, that per-action availability is reported via capabilities, and how progress is signalled. This is strong disclosure; it only lacks rate-limit/cost hints.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Front-loaded with the one-line purpose, then cleanly grouped into connectivity/telemetry/progress plus a secondary field list. It is long and partly overlaps the existing output schema, but nearly every clause carries semantic meaning (state interpretations) rather than filler.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

An output schema exists, so return values need not be explained, yet the description still adds interpretative value on states and capabilities. Combined with the auth note in the schema, an agent has almost everything needed; the only gap is executionId format/source.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 50%: authKey is fully documented in the schema, but executionId has no schema description and the text only says 'by executionId' without format or provenance. With half the parameters undocumented elsewhere, the description does not fully compensate.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

States a specific verb and resource with scope: 'One read of a job while it runs, by executionId.' The name, title, and description all agree, and the timing ('while it runs') plus the enumerated return facets make it distinguishable from siblings like getStatus, getArtifact, and listJobs.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines3/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

It gives timing context ('while it runs') but never names an alternative or an exclusion. An agent is left to infer that this is the live-inspection tool versus getStatus or listJobs rather than being routed explicitly to it.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

listJobsList your jobsA

Read-onlyIdempotent

Inspect

List the jobs THIS account has submitted, newest first — the tool to reach for when you do not have an executionId in hand (a fresh session, a lost id, "what am I still running?"). Each row carries executionId, submittedAtMs, the spec's image, budgetUsd, status, stage, artifactsReady, the same nextAction hint getStatus returns, and its deliveryOutcome, priceUsd and priceStatus; call getStatus for the full detail of any one of them. Returns count (rows served) and totalCount (jobs you have) — when they differ the list was truncated to the newest limit. Free, books nothing, and scoped: it never returns a job another account submitted.

ParametersJSON Schema

Name	Required	Description	Default
`limit`	No	How many of the newest jobs to return. Default 20, maximum 100; out-of-range values are clamped, never rejected.
`authKey`	No	Usually omit this. The customer door — the OAuth-signed-in HTTPS connector at https://symbioza.dev/mcp — authenticates the session itself, so no key is passed per call. Only a self-hosted or test transport passes a key here.

Output Schema

ParametersJSON Schema

Name	Required	Description
`code`	No	Machine-readable refusal code: unauthorized, invalid_spec, budget_too_low, insufficient_balance, access_required, secret_conflict, no_compute, unavailable, refused
`jobs`	No	This account's jobs, newest first
`count`	No
`action`	No	The one next step: authenticate, fix_spec, raise_budget, add_funds, claim_credit, add_access, relax_spec_or_retry, retry_later, contact_support
`reason`	No	Why, in a customer-safe sentence
`rejected`	No	Present and true when the call was refused; nothing was booked or charged
`totalCount`	No	All of this account's jobs; larger than count when the list is truncated
`supportedImages`	No	On an unverified-image refusal: the images the door accepts

TDQS

A4.9/5.0

Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations already cover readOnly/idempotent/non-destructive, and the description adds genuinely new behavior: it is free and books nothing, it is account-scoped so it never returns another account's job, and it discloses truncation semantics (count vs totalCount diverging means the list was clipped to newest limit). That is real context beyond the structured hints.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

A single dense paragraph that front-loads purpose, then use-case, then return shape, then cost/scope caveats. Every clause carries information and it ends on the two guarantees an agent most needs (free, account-scoped); no filler.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

An output schema exists and the description still enumerates the row fields plus count/totalCount, so the agent knows both the invocation contract and the shape of results. For a read-only listing tool with annotations and an output schema, nothing material is missing.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 100%, so limit and authKey are already documented (including clamping). The description nonetheless ties limit to observable behavior by explaining that count/totalCount divergence means truncation to the newest `limit`, adding meaning above the schema's baseline.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

States a specific verb+resource+scope: jobs THIS account submitted, newest first. It explicitly contrasts itself with getStatus ("call getStatus for the full detail of any one of them") and with the sibling family by naming executionId as the alternative entry point.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Gives the exact triggering conditions ("when you do not have an executionId in hand — a fresh session, a lost id, 'what am I still running?'") and routes to the alternative (getStatus) for per-job detail. Nothing is left to inference.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

listSecretsList your saved secrets (metadata only, never values)A

Read-onlyIdempotent

Inspect

List the secret sets saved on THIS account — metadata only (names, types and variable names); values are never returned, to anyone, by this or any other tool (a job that attaches a set is what receives them). Each row carries id, name, type (huggingface, object-storage, wandb or custom), provider (object storage only), fields (the environment variable NAMES the set injects) and status — metadata only, never a value, and only this account's sets. Attach a set to a job with the spec's secrets, by name (or by id). Credentials are saved by the user at https://symbioza.dev/app/secrets — never pasted into chat and never put in env. Call it again after the user says they saved a set: continue only if it is listed. Free, books nothing.

ParametersJSON Schema

Name	Required	Description	Default
No parameters

Output Schema

ParametersJSON Schema

Name	Required	Description
`code`	No	Machine-readable refusal code: unauthorized, invalid_spec, budget_too_low, insufficient_balance, access_required, secret_conflict, no_compute, unavailable, refused
`note`	No
`count`	No
`action`	No	The one next step: authenticate, fix_spec, raise_budget, add_funds, claim_credit, add_access, relax_spec_or_retry, retry_later, contact_support
`reason`	No	Why, in a customer-safe sentence
`secrets`	No	This account's saved secret sets: metadata only, never a value
`rejected`	No	Present and true when the call was refused; nothing was booked or charged
`manageUrl`	No	Where the user adds, rotates or deletes a saved set
`supportedImages`	No	On an unverified-image refusal: the images the door accepts

TDQS

A4.3/5.0

Behavior4/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations already declare readOnly, idempotent, closed-world and non-destructive, so the safety profile is covered. The description adds genuinely useful context beyond that: values are never returned by this or any other tool, credentials are created out-of-band at a URL rather than passed in chat or env, and the call is free and books nothing. This is meaningful extra behavioral disclosure.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Front-loaded with the core action and scope, and most sentences carry distinct information (row fields, attach path, credential URL, re-check flow). It repeats 'metadata only, never a value' roughly three times, which costs some density but does not bury the lead.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

With an output schema present the description need not explain return values, and annotations cover the safety profile, so what remains is workflow context — which it supplies (how to attach a set, where credentials come from, verifying after creation). Complete enough to call correctly; only the repetition is a minor blemish.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

The tool takes zero parameters, so there is no parameter semantics to add and the baseline is 4. The description instead enumerates the returned row fields, which is beyond the burden of a parameterless tool.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description names a specific verb and resource (list the secret sets saved on THIS account) and immediately bounds the scope to metadata only, distinguishing it from any tool that would return values. An agent knows exactly what it gets back (id, name, type, provider, fields, status) without opening the schema.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

It gives real usage context: attach a set to a job via the spec's secrets, and re-call after the user reports saving a set, continuing only if it is listed. No sibling alternative exists to route against, so the lack of explicit 'use X instead' is not a gap, but it stops short of a full when/when-not treatment.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

submitJobRun a job on a rented machine (up to your spending limit)AInspect

Estimate first — budgetUsd is optional there — then ask the user for budgetUsd (never choose one for them) and call submitJob only after the user has approved the estimate at that budget. Run a containerized workload on a rented cloud machine sized to the spec — GPU, and as much disk as minDiskGb asks for, so the local machine's disk and bandwidth are irrelevant (the machine downloads datasets itself). Queues the job and returns a durable executionId immediately (status "queued"); Symbioza provisions, runs, collects the output files and tears down, all under the budgetUsd ceiling. Write every deliverable to /workspace/artifacts/ — only that directory is collected and returned by getArtifact; a command that saves its model to the working directory or any other path finishes with nothing delivered. image must be one of the images we have verified launch on our machines: pytorch/pytorch:2.4.1-cuda12.1-cudnn9-runtime, pytorch/pytorch:2.4.0-cuda12.1-cudnn9-runtime, tensorflow/tensorflow:2.15.0. Any other image is refused with code invalid_spec, action fix_spec and a supportedImages list, before anything is booked or charged. budgetUsd is what you are willing to spend on the job, and you are never billed more than that. Call estimateExecution first if you have not priced this spec — it also tells you whether this account can cover the budget, and whether it holds free compute credit still to claim. A refusal here carries a machine-readable code (unauthorized, invalid_spec, budget_too_low, insufficient_balance, access_required, secret_conflict, no_compute, unavailable, refused) and one action (authenticate, fix_spec, raise_budget, add_funds, claim_credit, add_access, relax_spec_or_retry, retry_later, contact_support), so branch on those rather than on the sentence; a refusal books no machine and charges nothing. insufficient_balance with action claim_credit means free credit is waiting on the account — claim it at https://symbioza.dev/app and resubmit; insufficient_balance also carries shortfallUsd and topupUrl. access_required and secret_conflict carry the same access block estimateExecution returns, with a Symbioza setup link for each missing saved secret: give the user that link, and after they say it is saved call listSecrets and submit again. On acceptance each attached set is pinned to its id. OPTIONAL: if your workload can count its own work, write /workspace/artifacts/.symbiosa/metrics.json with {"unit": "configuration", "units_planned": 100, "units_completed": 100, "elapsed_seconds": 4821} — unit is free text naming what you counted, partial progress is valid evidence on a failed run, and the whole file is optional (omit it and throughput is simply recorded as UNKNOWN; do not invent a unit to fill it in). It is recorded as observed evidence about your workload and never affects what you are charged; a missing, malformed or incoherent file changes nothing about the run. You may disconnect and poll getStatus later. If an attempt fails and the job is moved to a replacement machine, the newest checkpoint streamed off the previous box (files named ckpt_step.pt, N = digits) is pushed onto the new box under /workspace/.symbiosa/resume/ (outside the artifacts directory) and SYMBIOSA_RESUME is set to its absolute path there — your command must load that file to continue, or the attempt restarts from the beginning. The variable is unset on a first attempt, when no checkpoint was kept, when kept checkpoints span more than one directory (multi-arm layouts are not restored), when checkpoint.enabled is false, when the box lacks disk headroom, or when the push failed; restore is best-effort salvage and never fails a run. Work since the last streamed checkpoint is lost, and the push's hash check proves the transfer, not that the checkpoint itself is intact. Only when the user explicitly requires a particular model for their own benchmark, set gpu.gpuNameExact: true alongside gpu.gpuName to pin the CARD MODEL: only that model is booked and every substitute is refused, including a faster one, because when the measured quantity is the hardware a better card is as wrong as a slower one; it holds the model across every run. pinHost is accepted and ignored. The same model is sold at different board powers under one name, so for an energy or efficiency measurement also declare gpu.minPowerLimitW (watts): hosts below it are refused, and a host publishing no power figure is refused rather than assumed un-throttled. Declare gpu.minCudaVersion (e.g. 13.0) when your image pins a torch/flash-attn build against a CUDA version: only hosts whose driver advertises at least that CUDA are booked, and a host that publishes no CUDA ceiling is refused rather than booked. Declare datasetGb as the total data the box will ingest, including data it downloads itself (not only what datasets[] stages), so a host too slow to pull it in time is filtered before booking. With a clientRequestId, the same key with the same spec returns the same job, marked duplicate: true, instead of booking and charging for a second one; the same key with a different spec is refused with invalid_spec. Without a key, an identical submission (same spec and budget) from the same account returns the same job marked duplicate: true and deduplicated: true and books nothing while that job is still queued or running, at any time after it failed, was cancelled or was lost (then the response carries a hint instead of starting a second paid attempt), and within 10 minutes of its acceptance if it delivered. Keep one clientRequestId for one submission across every resend. To run a failed, cancelled or lost job again, call submitJob with retryOf set to its executionId and a new clientRequestId: that starts a fresh attempt at once, checked against the budget and balance again, even with the same spec, and its response names the earlier job as retryOf; a retryOf that is not a failed, cancelled or lost job of this account is refused with invalid_spec. A new clientRequestId always starts a new job. Pass the specDigest estimateExecution returned and a spec or budget changed since that estimate is refused with invalid_spec, so what runs is what you were quoted.

ParametersJSON Schema

Name	Required	Description
`spec`	Yes	The job spec: image + command run on a rented box sized by minDiskGb (and gpu, when declared), under budgetUsd. Supply the workload: the image, the command, its inputs, the expected outputs and budgetUsd. Symbioza sizes the machine — do not ask the user to choose CPU cores, RAM, VRAM, a GPU model, a provider or a region, and do not derive a resource minimum from a thread count, a worker count or a measurement taken somewhere else. gpu.minVramGb, minRamGb and minCpuCores are optional expert floors: declare one only when the workload has evidence for it, and omit it otherwise — an omitted floor is sized by Symbioza, never treated as zero. The gpu block is optional: leave it out and the job is still GPU work with Symbioza sizing the card; declare it to carry minCudaVersion or an evidenced floor.
`authKey`	No	Usually omit this. The customer door — the OAuth-signed-in HTTPS connector at https://symbioza.dev/mcp — authenticates the session itself, so no key is passed per call. Only a self-hosted or test transport passes a key here.
`retryOf`	No	Only to run a failed, cancelled or lost job again, on purpose: that job's executionId, sent with a new clientRequestId. The retry starts at once as a new job, checked against the budget and balance again, may carry the same spec, and its response names the earlier job as retryOf. Refused with invalid_spec when the job is not this account's, was delivered, or is still queued or running.
`budgetUsd`	No	Optional cross-check: must EQUAL spec.budgetUsd or the submit is rejected. Never overrides it.
`specDigest`	No	Optional, recommended: the specDigest estimateExecution returned for this spec — if the spec or its budgetUsd changed since that estimate, the submit is refused with invalid_spec and nothing is booked or charged.
`clientRequestId`	No	Optional, recommended: a key of your own for this submission, 1 to 128 printable ASCII characters — keep it for every resend of the same submission (after a timeout or a lost response): the same key with the same spec returns the same job whatever its outcome: the same executionId with duplicate: true, nothing booked or charged twice; the same key with a different spec is refused with invalid_spec, so use a new key for a new spec or for a retryOf retry.

Output Schema

ParametersJSON Schema

Name	Required	Description
`code`	No	Machine-readable refusal code: unauthorized, invalid_spec, budget_too_low, insufficient_balance, access_required, secret_conflict, no_compute, unavailable, refused
`hint`	No	On a repeated submission whose job ended without delivering (failed, cancelled or lost): nothing new was booked, and this says how to run it again — submitJob with retryOf and a new clientRequestId
`access`	No	On access_required or secret_conflict: what the spec's secrets need, with setup links
`action`	No	The one next step: authenticate, fix_spec, raise_budget, add_funds, claim_credit, add_access, relax_spec_or_retry, retry_later, contact_support
`reason`	No	Why, in a customer-safe sentence
`status`	No	"queued" on acceptance
`retryOf`	No	The earlier job this one retries: set when submitJob was called with retryOf, on the retry and on every resend of it
`rejected`	No	Present and true when the call was refused; nothing was booked or charged
`topupUrl`	No	On insufficient_balance: the link to give the user for that credit
`duplicate`	No	true when a repeated submit returned the job the first call started: a clientRequestId retry, or an identical keyless submission
`executionId`	No	Your durable job id
`deduplicated`	No	true when no clientRequestId was sent and an identical submission from this account (still queued or running, ended without delivering at any age, or delivered and accepted within the last 10 minutes) was matched; nothing was booked
`shortfallUsd`	No	On insufficient_balance: whole dollars of credit this job still needs
`supportedImages`	No	On an unverified-image refusal: the images the door accepts

TDQS

A4.4/5.0

Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

With annotations already declaring readOnlyHint=false, openWorldHint=true, idempotentHint=false, and destructiveHint=false, the description still adds substantial behavioral context beyond them: immediate queuing and durable executionId, budget ceiling enforcement, artifact collection only from /workspace/artifacts/, image allowlist refusals, refusal codes and actions, checkpoint restore semantics, deduplication rules, and that a refusal books nothing and charges nothing. It does not contradict any annotation.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness2/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is extremely long and reads as a wall of text. While it front-loads the critical estimate-then-submit workflow and many sentences carry real information, it repeats schema descriptions and covers numerous edge cases at length, making it poorly sized for an agent to scan. Conciseness suffers despite the useful content.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Given the tool's complexity, the rich input schema, the annotations, and the presence of an output schema, the description is more than complete. It covers the entire lifecycle: estimate, approval, budget, submission, refusal codes, artifact location, checkpoint resume, retry, deduplication, and secret access. Nothing material is missing for an agent to invoke the tool correctly.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema description coverage is 100%, so the baseline is 3. The description adds some cross-parameter context, such as clientRequestId deduplication, retryOf semantics, specDigest enforcement, and budgetUsd cross-checking, but the schema already documents these parameters in exhaustive detail. Overall it does not add enough meaning beyond the schema to merit a 4.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description states a specific verb and resource: running a containerized workload on a rented cloud machine sized to the spec, queuing it, returning an executionId, and provisioning/teardown under budgetUsd. It distinguishes this from siblings by naming estimateExecution, getStatus, getArtifact, listSecrets, and retryOf behavior. An agent can tell exactly what the tool does and how it differs from alternatives.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

It gives explicit sequencing and alternatives: call estimateExecution first if the spec has not been priced, ask the user for budgetUsd, and only call submitJob after the user approves the estimate. It also names when to use retryOf versus a new clientRequestId, and when to branch on refusal codes rather than retrying blindly. This is explicit when-to-use and when-not-to-use guidance.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

1 tool updateOct 9, 2026
- AddedinspectRun

2 tool updatesOct 7, 2026

ChangedestimateExecution1 field changed

changedInput schema / properties / spec / properties / pinHost / description

Previous value: -"Set true for a measurement/benchmark workload whose numbers are only comparable on one machine. The job then FAILS if its box is lost mid-run instead of being rerouted to a different one — a silent host swap would make the results meaningless. Default false: a lost box may be replaced by another machine."New value: +"Accepted for compatibility and ignored. A machine that is lost or never starts may be replaced by another one, inside budgetUsd. To keep a benchmark on one GPU model, set gpu.gpuName with gpu.gpuNameExact."

ChangedsubmitJob1 field changed

changedInput schema / properties / spec / properties / pinHost / description

Previous value: -"Set true for a measurement/benchmark workload whose numbers are only comparable on one machine. The job then FAILS if its box is lost mid-run instead of being rerouted to a different one — a silent host swap would make the results meaningless. Default false: a lost box may be replaced by another machine."New value: +"Accepted for compatibility and ignored. A machine that is lost or never starts may be replaced by another one, inside budgetUsd. To keep a benchmark on one GPU model, set gpu.gpuName with gpu.gpuNameExact."

3 tool updatesOct 7, 2026

ChangedgetArtifact2 fields changed

addedOutput schema / properties / executedAttempt

Added value: +{
+  "anyOf": [
+    {
+      "additionalProperties": {},
+      "properties": {
+        "exitCode": {
+          "anyOf": [
+            {
+              "type": "number"
+            },
+            {
+              "type": "null"
+            }
+          ]
+        },
+        "exitedAtMs": {
+          "anyOf": [
+            {
+              "type": "number"
+            },
+            {
+              "type": "null"
+            }
+          ]
+        },
+        "index": {
+          "type": "number"
+        },
+        "startedAtMs": {
+          "anyOf": [
+            {
+              "type": "number"
+            },
+            {
+              "type": "null"
+            }
+          ]
+        }
+      },
+      "required": [
+        "index",
+        "exitCode",
+        "startedAtMs",
+        "exitedAtMs"
+      ],
+      "type": "object"
+    },
+    {
+      "type": "null"
+    }
+  ],
+  "description": "The attempt that ran your command (0-based); null when no attempt ran it"
+}

addedOutput schema / properties / recoveryAssessment

Added value: +{
+  "anyOf": [
+    {
+      "additionalProperties": {},
+      "properties": {
+        "automatic": {
+          "type": "boolean"
+        },
+        "blocker": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "description": "dependency_access | constraint | budget | capacity | checkpoint | cleanup_unconfirmed | unknown; null unless blocked"
+        },
+        "mode": {
+          "description": "none | resume | restart | blocked | user_stopped",
+          "type": "string"
+        },
+        "needsNewApproval": {
+          "type": "boolean"
+        },
+        "nextAction": {
+          "type": "string"
+        },
+        "preserved": {
+          "additionalProperties": {},
+          "properties": {
+            "checkpoint": {
+              "type": "boolean"
+            },
+            "files": {
+              "type": "number"
+            },
+            "verifiedFiles": {
+              "type": "number"
+            }
+          },
+          "required": [
+            "files",
+            "verifiedFiles",
+            "checkpoint"
+          ],
+          "type": "object"
+        },
+        "reason": {
+          "type": "string"
+        }
+      },
+      "required": [
+        "mode",
+        "blocker",
+        "reason",
+        "nextAction",
+        "preserved",
+        "needsNewApproval",
+        "automatic"
+      ],
+      "type": "object"
+    },
+    {
+      "type": "null"
+    }
+  ]
+}

ChangedgetStatus3 fields changed

addedOutput schema / properties / executedAttempt

Added value: +{
+  "anyOf": [
+    {
+      "additionalProperties": {},
+      "properties": {
+        "exitCode": {
+          "anyOf": [
+            {
+              "type": "number"
+            },
+            {
+              "type": "null"
+            }
+          ]
+        },
+        "exitedAtMs": {
+          "anyOf": [
+            {
+              "type": "number"
+            },
+            {
+              "type": "null"
+            }
+          ]
+        },
+        "index": {
+          "type": "number"
+        },
+        "startedAtMs": {
+          "anyOf": [
+            {
+              "type": "number"
+            },
+            {
+              "type": "null"
+            }
+          ]
+        }
+      },
+      "required": [
+        "index",
+        "exitCode",
+        "startedAtMs",
+        "exitedAtMs"
+      ],
+      "type": "object"
+    },
+    {
+      "type": "null"
+    }
+  ],
+  "description": "The attempt that ran your command (0-based); null when no attempt ran it"
+}

addedOutput schema / properties / machinesBooked

Added value: +{
+  "anyOf": [
+    {
+      "type": "number"
+    },
+    {
+      "type": "null"
+    }
+  ],
+  "description": "Distinct machines this job held; attempts also counts any refused before a machine was booked"
+}

addedOutput schema / properties / recoveryAssessment

Added value: +{
+  "anyOf": [
+    {
+      "additionalProperties": {},
+      "properties": {
+        "automatic": {
+          "type": "boolean"
+        },
+        "blocker": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "description": "dependency_access | constraint | budget | capacity | checkpoint | cleanup_unconfirmed | unknown; null unless blocked"
+        },
+        "mode": {
+          "description": "none | resume | restart | blocked | user_stopped",
+          "type": "string"
+        },
+        "needsNewApproval": {
+          "type": "boolean"
+        },
+        "nextAction": {
+          "type": "string"
+        },
+        "preserved": {
+          "additionalProperties": {},
+          "properties": {
+            "checkpoint": {
+              "type": "boolean"
+            },
+            "files": {
+              "type": "number"
+            },
+            "verifiedFiles": {
+              "type": "number"
+            }
+          },
+          "required": [
+            "files",
+            "verifiedFiles",
+            "checkpoint"
+          ],
+          "type": "object"
+        },
+        "reason": {
+          "type": "string"
+        }
+      },
+      "required": [
+        "mode",
+        "blocker",
+        "reason",
+        "nextAction",
+        "preserved",
+        "needsNewApproval",
+        "automatic"
+      ],
+      "type": "object"
+    },
+    {
+      "type": "null"
+    }
+  ]
+}

ChangedlistJobs3 fields changed

addedOutput schema / properties / jobs / items / properties / executedAttempt

Added value: +{
+  "anyOf": [
+    {
+      "additionalProperties": {},
+      "properties": {
+        "exitCode": {
+          "anyOf": [
+            {
+              "type": "number"
+            },
+            {
+              "type": "null"
+            }
+          ]
+        },
+        "exitedAtMs": {
+          "anyOf": [
+            {
+              "type": "number"
+            },
+            {
+              "type": "null"
+            }
+          ]
+        },
+        "index": {
+          "type": "number"
+        },
+        "startedAtMs": {
+          "anyOf": [
+            {
+              "type": "number"
+            },
+            {
+              "type": "null"
+            }
+          ]
+        }
+      },
+      "required": [
+        "index",
+        "exitCode",
+        "startedAtMs",
+        "exitedAtMs"
+      ],
+      "type": "object"
+    },
+    {
+      "type": "null"
+    }
+  ],
+  "description": "The attempt that ran your command (0-based); null when no attempt ran it"
+}

addedOutput schema / properties / jobs / items / properties / exitCode

Added value: +{
+  "anyOf": [
+    {
+      "type": "number"
+    },
+    {
+      "type": "null"
+    }
+  ]
+}

addedOutput schema / properties / jobs / items / properties / recoveryAssessment

Added value: +{
+  "anyOf": [
+    {
+      "additionalProperties": {},
+      "properties": {
+        "automatic": {
+          "type": "boolean"
+        },
+        "blocker": {
+          "anyOf": [
+            {
+              "type": "string"
+            },
+            {
+              "type": "null"
+            }
+          ],
+          "description": "dependency_access | constraint | budget | capacity | checkpoint | cleanup_unconfirmed | unknown; null unless blocked"
+        },
+        "mode": {
+          "description": "none | resume | restart | blocked | user_stopped",
+          "type": "string"
+        },
+        "needsNewApproval": {
+          "type": "boolean"
+        },
+        "nextAction": {
+          "type": "string"
+        },
+        "preserved": {
+          "additionalProperties": {},
+          "properties": {
+            "checkpoint": {
+              "type": "boolean"
+            },
+            "files": {
+              "type": "number"
+            },
+            "verifiedFiles": {
+              "type": "number"
+            }
+          },
+          "required": [
+            "files",
+            "verifiedFiles",
+            "checkpoint"
+          ],
+          "type": "object"
+        },
+        "reason": {
+          "type": "string"
+        }
+      },
+      "required": [
+        "mode",
+        "blocker",
+        "reason",
+        "nextAction",
+        "preserved",
+        "needsNewApproval",
+        "automatic"
+      ],
+      "type": "object"
+    },
+    {
+      "type": "null"
+    }
+  ]
+}

1 tool updateOct 7, 2026

ChangedestimateExecution2 fields changed

changedOutput schema / properties / estimatedPriceUsd / description

Previous value: -"What this spec is expected to bill you, in dollars — present only when price.evidence is customer (priced at your own maxRuntimeSeconds); absent when nothing about this job predicts its runtime"New value: +"What this spec is expected to bill you, in dollars — present only when price.evidence is customer (the estimate priced at your own maxRuntimeSeconds), and then equal to estimate.likelyUsd; otherwise absent and the total is estimate.likelyUsd"

changedOutput schema / properties / runtime / description

Previous value: -"basis, referenceSeconds, boundSeconds, boundBy, note: the runtime the estimate assumes and where the run would be stopped"New value: +"boundSeconds, boundBy (budget | customer | platform), note: where the run would be STOPPED on that machine — a limit, never a prediction. The expected runtime is estimate.runtime"

1 tool updateOct 7, 2026

ChangedestimateExecution2 fields changed

addedOutput schema / properties / estimate

Added value: +{
+  "additionalProperties": {},
+  "description": "The best-effort estimate of the total bill: show likelyUsd with its range, confidence and disclaimer, or perHourUsd when rate_only. budgetUsd stays the hard cap",
+  "properties": {
+    "assumptions": {
+      "items": {
+        "type": "string"
+      },
+      "type": "array"
+    },
+    "confidence": {
+      "anyOf": [
+        {
+          "type": "string"
+        },
+        {
+          "type": "null"
+        }
+      ],
+      "description": "low | medium | high — how strong the runtime evidence is; null when rate_only"
+    },
+    "costBreakdown": {
+      "description": "The likely total line by line, in the bill's own categories; the lines sum to likelyUsd",
+      "items": {
+        "additionalProperties": {},
+        "properties": {
+          "category": {
+            "type": "string"
+          },
+          "label": {
+            "type": "string"
+          },
+          "usd": {
+            "type": "number"
+          }
+        },
+        "required": [
+          "category",
+          "label",
+          "usd"
+        ],
+        "type": "object"
+      },
+      "type": "array"
+    },
+    "disclaimer": {
+      "description": "Show it with the figures: best-effort, may differ, budgetUsd is the hard maximum",
+      "type": "string"
+    },
+    "highUsd": {
+      "anyOf": [
+        {
+          "type": "number"
+        },
+        {
+          "type": "null"
+        }
+      ],
+      "description": "The range the charge likely falls in; never above budgetUsd"
+    },
+    "likelyUsd": {
+      "anyOf": [
+        {
+          "type": "number"
+        },
+        {
+          "type": "null"
+        }
+      ],
+      "description": "The likely total charge for a run that works, in dollars; a best-effort estimate, never a quote. null when rate_only"
+    },
+    "limitedBySpendingLimit": {
+      "description": "true when the top of the range is budgetUsd itself: the run is stopped there and billed no more",
+      "type": "boolean"
+    },
+    "lowUsd": {
+      "anyOf": [
+        {
+          "type": "number"
+        },
+        {
+          "type": "null"
+        }
+      ]
+    },
+    "perHourUsd": {
+      "anyOf": [
+        {
+          "type": "number"
+        },
+        {
+          "type": "null"
+        }
+      ],
+      "description": "What each hour of running costs on the machine Symbioza would book"
+    },
+    "runtime": {
+      "additionalProperties": {},
+      "description": "likelySeconds, lowSeconds, highSeconds, basis (same-workload | workload-family | command-heuristic | platform-history | customer-cap | none), evidence (one sentence for the user), samples",
+      "properties": {},
+      "type": "object"
+    },
+    "status": {
+      "description": "best_effort: likelyUsd/lowUsd/highUsd are set | rate_only: nothing predicts the runtime, so there is no total — show perHourUsd",
+      "type": "string"
+    }
+  },
+  "required": [
+    "status",
+    "likelyUsd",
+    "lowUsd",
+    "highUsd",
+    "confidence",
+    "perHourUsd",
+    "runtime",
+    "costBreakdown",
+    "limitedBySpendingLimit",
+    "assumptions",
+    "disclaimer"
+  ],
+  "type": "object"
+}

changedOutput schema / properties / price / properties / evidence / description

Previous value: -"customer: priced at your own maxRuntimeSeconds | insufficient: nothing about this job predicts its runtime, so this is not a quote"New value: +"customer: priced at your own maxRuntimeSeconds | best-effort: a likely total and range in estimate, not a quote | insufficient: nothing about this job predicts its runtime, so only the hourly price"

1 tool updateOct 5, 2026
- ChangedlistSecrets17 fields changed
  - removedInput schema / properties / authKey
```
Removed value: -{
-  "description": "Usually omit this. The customer door — the OAuth-signed-in HTTPS connector at https://symbioza.dev/mcp — authenticates the session itself, so no key is passed per call. Only a self-hosted or test transport passes a key here.",
-  "type": "string"
-}
```
  - changedOutput schema / additionalProperties
```
Previous value: -{}New value: +false
```
  - addedOutput schema / properties / count / maximum
```
Added value: +9007199254740991
```
  - addedOutput schema / properties / count / minimum
```
Added value: +0
```
  - changedOutput schema / properties / count / type
```
Previous value: -"number"New value: +"integer"
```
  - changedOutput schema / properties / secrets / description
```
Previous value: -"This account's saved secret sets: names and types only"New value: +"This account's saved secret sets: metadata only, never a value"
```
  - changedOutput schema / properties / secrets / items / additionalProperties
```
Previous value: -{}New value: +false
```
  - addedOutput schema / properties / secrets / items / properties / fields / items / pattern
```
Added value: +"^[A-Za-z_][A-Za-z0-9_]{0,127}$"
```
  - addedOutput schema / properties / secrets / items / properties / id / description
```
Added value: +"The set's id (sec_…); spec.secrets accepts it in place of the name"
```
  - addedOutput schema / properties / secrets / items / properties / id / pattern
```
Added value: +"^sec_[A-Za-z0-9_-]{22}$"
```
  - removedOutput schema / properties / secrets / items / properties / lastUsedAtMs
```
Removed value: -{
-  "type": "number"
-}
```
  - addedOutput schema / properties / secrets / items / properties / name / description
```
Added value: +"The set's name: what spec.secrets references"
```
  - addedOutput schema / properties / secrets / items / properties / name / pattern
```
Added value: +"^[a-z0-9][a-z0-9_-]{0,62}$"
```
  - addedOutput schema / properties / secrets / items / properties / provider / description
```
Added value: +"Object storage only: which kind"
```
  - addedOutput schema / properties / secrets / items / properties / provider / enum
```
Added value: +[
+  "s3",
+  "r2",
+  "s3-compatible"
+]
```
  - addedOutput schema / properties / secrets / items / properties / status / const
```
Added value: +"ready"
```
  - addedOutput schema / properties / secrets / items / properties / type / enum
```
Added value: +[
+  "huggingface",
+  "object-storage",
+  "wandb",
+  "custom"
+]
```

2 tool updatesOct 3, 2026

ChangedestimateExecution1 field changed

changedInput schema / properties / spec / properties / gpu / properties / count / description

Previous value: -"GPUs on ONE machine. Must be 1 — a higher count is REFUSED, never silently reduced to one card. Omit it."New value: +"Exact number of GPUs on ONE machine (never N separate machines). Omit it for one GPU, or declare 2; any other count is REFUSED, never silently changed. Declare it only when the workload runs that many processes on a single host (for example torchrun --nproc_per_node=N); minVramGb then applies to each GPU."

ChangedsubmitJob1 field changed

changedInput schema / properties / spec / properties / gpu / properties / count / description

Previous value: -"GPUs on ONE machine. Must be 1 — a higher count is REFUSED, never silently reduced to one card. Omit it."New value: +"Exact number of GPUs on ONE machine (never N separate machines). Omit it for one GPU, or declare 2; any other count is REFUSED, never silently changed. Declare it only when the workload runs that many processes on a single host (for example torchrun --nproc_per_node=N); minVramGb then applies to each GPU."

2 tool updatesOct 1, 2026

ChangedestimateExecution3 fields changed

changedInput schema / properties / spec / properties / gpu / properties / count / description

Previous value: -"GPUs on ONE machine. Must be 1 — a higher count is REFUSED, never silently reduced to one card. Omit it unless you mean several cards on a single host (that is not N separate machines)."New value: +"GPUs on ONE machine. Must be 1 — a higher count is REFUSED, never silently reduced to one card. Omit it."

changedInput schema / properties / spec / properties / gpu / properties / count / maximum
```
Previous value: -9007199254740991New value: +16
```

changedInput schema / properties / spec / properties / gpu / properties / minVramGb / description

Previous value: -"Optional. A VRAM floor the box must meet, only when the workload has an evidenced one (a measured peak, a known model footprint). Omit it when unknown: Symbioza then sizes the GPU itself. Never ask the user to choose a tier, and never derive a number from an unrelated run."New value: +"Optional. A VRAM floor the box must meet, PER GPU, only when the workload has an evidenced one (a measured peak, a known model footprint). Omit it when unknown: Symbioza then sizes the GPU itself. Never ask the user to choose a tier, and never derive a number from an unrelated run."

ChangedsubmitJob3 fields changed

changedInput schema / properties / spec / properties / gpu / properties / count / description

Previous value: -"GPUs on ONE machine. Must be 1 — a higher count is REFUSED, never silently reduced to one card. Omit it unless you mean several cards on a single host (that is not N separate machines)."New value: +"GPUs on ONE machine. Must be 1 — a higher count is REFUSED, never silently reduced to one card. Omit it."

changedInput schema / properties / spec / properties / gpu / properties / count / maximum
```
Previous value: -9007199254740991New value: +16
```

changedInput schema / properties / spec / properties / gpu / properties / minVramGb / description

Previous value: -"Optional. A VRAM floor the box must meet, only when the workload has an evidenced one (a measured peak, a known model footprint). Omit it when unknown: Symbioza then sizes the GPU itself. Never ask the user to choose a tier, and never derive a number from an unrelated run."New value: +"Optional. A VRAM floor the box must meet, PER GPU, only when the workload has an evidenced one (a measured peak, a known model footprint). Omit it when unknown: Symbioza then sizes the GPU itself. Never ask the user to choose a tier, and never derive a number from an unrelated run."

1 tool updateSep 30, 2026

ChangedestimateExecution14 fields changed

changedInput schema / properties / spec / description

Previous value: -"The job spec: image + command run on a rented box sized by minDiskGb (and gpu, when declared), under budgetUsd. Supply the workload: the image, the command, its inputs, the expected outputs and budgetUsd. Symbioza sizes the machine — do not ask the user to choose CPU cores, RAM, VRAM, a GPU model, a provider or a region, and do not derive a resource minimum from a thread count, a worker count or a measurement taken somewhere else. gpu.minVramGb, minRamGb and minCpuCores are optional expert floors: declare one only when the workload has evidence for it, and omit it otherwise — an omitted floor is sized by Symbioza, never treated as zero. The gpu block is optional: leave it out and the job is still GPU work with Symbioza sizing the card; declare it to carry minCudaVersion or an evidenced floor."New value: +"The job spec to estimate: image + command run on a rented box sized by minDiskGb (and gpu, when declared). budgetUsd is optional here — leave it out to see the likely cost before the user chooses one; submitJob requires it. Supply the workload: the image, the command, its inputs, the expected outputs and budgetUsd. Symbioza sizes the machine — do not ask the user to choose CPU cores, RAM, VRAM, a GPU model, a provider or a region, and do not derive a resource minimum from a thread count, a worker count or a measurement taken somewhere else. gpu.minVramGb, minRamGb and minCpuCores are optional expert floors: declare one only when the workload has evidence for it, and omit it otherwise — an omitted floor is sized by Symbioza, never treated as zero. The gpu block is optional: leave it out and the job is still GPU work with Symbioza sizing the card; declare it to carry minCudaVersion or an evidenced floor."

changedInput schema / properties / spec / properties / budgetUsd / description

Previous value: -"Hard spend ceiling for the whole job, USD. The job is refused or stopped rather than exceeding it."New value: +"Optional here: the job's hard spend ceiling, USD. Leave it out to see the estimate before the user chooses one; submitJob requires it, so estimate again with the user's budget before submitting."

changedInput schema / properties / spec / required

Previous value: -[
-  "image",
-  "command",
-  "budgetUsd"
-]New value: +[
+  "image",
+  "command"
+]

addedOutput schema / properties / account / properties / coversThisJob / anyOf

Added value: +[
+  {
+    "type": "boolean"
+  },
+  {
+    "type": "null"
+  }
+]

addedOutput schema / properties / account / properties / coversThisJob / description

Added value: +"Would submitJob accept this spec's budget today; null when the spec has no budgetUsd"

removedOutput schema / properties / account / properties / coversThisJob / type
```
Removed value: -"boolean"
```
addedOutput schema / properties / account / properties / topupUrl / description
```
Added value: +"With a budget: the link for the credit this job needs"
```

changedOutput schema / properties / account / required

Previous value: -[
-  "availableUsd",
-  "promoUsd",
-  "promoExpiresAtMs",
-  "coversThisJob",
-  "promoJobCapUsd",
-  "claimableUsd",
-  "nextAction",
-  "claimUrl",
-  "topupUrl"
-]New value: +[
+  "availableUsd",
+  "promoUsd",
+  "promoExpiresAtMs",
+  "coversThisJob",
+  "promoJobCapUsd",
+  "claimableUsd",
+  "nextAction",
+  "claimUrl"
+]

changedOutput schema / properties / budgetUsd / description

Previous value: -"Your own cap, read back; nothing bills past it"New value: +"Your own cap, read back; nothing bills past it. Absent when the spec has none"

changedOutput schema / properties / estimatedPriceUsd / description

Previous value: -"What this spec is expected to bill you, in dollars"New value: +"What this spec is expected to bill you, in dollars — present only when price.evidence is customer (priced at your own maxRuntimeSeconds); absent when nothing about this job predicts its runtime"

addedOutput schema / properties / price

Added value: +{
+  "additionalProperties": {},
+  "description": "What the price rests on, what it includes and what is still unknown",
+  "properties": {
+    "evidence": {
+      "description": "customer: priced at your own maxRuntimeSeconds | insufficient: nothing about this job predicts its runtime, so this is not a quote",
+      "type": "string"
+    },
+    "includes": {
+      "description": "What the estimate counts",
+      "items": {
+        "type": "string"
+      },
+      "type": "array"
+    },
+    "nextStep": {
+      "anyOf": [
+        {
+          "type": "string"
+        },
+        {
+          "type": "null"
+        }
+      ]
+    },
+    "perHourUsd": {
+      "anyOf": [
+        {
+          "type": "number"
+        },
+        {
+          "type": "null"
+        }
+      ],
+      "description": "What each hour of running costs on the machine Symbioza would book"
+    },
+    "summary": {
+      "description": "One sentence for the user",
+      "type": "string"
+    },
+    "unknown": {
+      "description": "What it cannot count yet",
+      "items": {
+        "type": "string"
+      },
+      "type": "array"
+    }
+  },
+  "required": [
+    "evidence",
+    "summary",
+    "perHourUsd",
+    "includes",
+    "unknown",
+    "nextStep"
+  ],
+  "type": "object"
+}

addedOutput schema / properties / readiness

Added value: +{
+  "additionalProperties": {},
+  "description": "What stands between this estimate and a submittable one, plus advice about its inputs",
+  "properties": {
+    "missing": {
+      "description": "budget_required | inputs_unverified | inputs_not_declared | access_required, each with a detail; the input notes are advice and never block a submit",
+      "items": {
+        "additionalProperties": {},
+        "properties": {
+          "code": {
+            "type": "string"
+          },
+          "detail": {
+            "type": "string"
+          }
+        },
+        "required": [
+          "code",
+          "detail"
+        ],
+        "type": "object"
+      },
+      "type": "array"
+    },
+    "status": {
+      "description": "preliminary while budget_required or access_required is listed; otherwise ready",
+      "type": "string"
+    }
+  },
+  "required": [
+    "status",
+    "missing"
+  ],
+  "type": "object"
+}

changedOutput schema / properties / runtime / description

Previous value: -"basis, referenceSeconds, boundSeconds, boundBy, note: the runtime the price assumes and where the run would be stopped"New value: +"basis, referenceSeconds, boundSeconds, boundBy, note: the runtime the estimate assumes and where the run would be stopped"

changedOutput schema / properties / specDigest / description

Previous value: -"Fingerprint of the exact spec and budget priced here; pass it to submitJob"New value: +"Fingerprint of the exact spec and budget priced here; pass it to submitJob. Absent when the spec has no budgetUsd: submitJob requires one"

5 tool updatesSep 30, 2026

ChangeddescribeDataset2 fields changed

changedOutput schema / properties / action / description

Previous value: -"The one next step: authenticate, fix_spec, raise_budget, add_funds, claim_credit, relax_spec_or_retry, retry_later, contact_support"New value: +"The one next step: authenticate, fix_spec, raise_budget, add_funds, claim_credit, add_access, relax_spec_or_retry, retry_later, contact_support"

changedOutput schema / properties / code / description

Previous value: -"Machine-readable refusal code: unauthorized, invalid_spec, budget_too_low, insufficient_balance, no_compute, unavailable, refused"New value: +"Machine-readable refusal code: unauthorized, invalid_spec, budget_too_low, insufficient_balance, access_required, secret_conflict, no_compute, unavailable, refused"

ChangedestimateExecution6 fields changed

changedInput schema / properties / spec / properties / env / description

Previous value: -"Environment variables for the container. They are stored with the job spec in clear on our control plane and are included in our routine backups, so do not put credentials or other secrets here — pass them in at run time or fetch them from your own store inside the command."New value: +"Non-sensitive configuration for the container (e.g. MODEL_NAME, EPOCHS). They are stored with the job spec in clear on our control plane and are included in our routine backups, so NEVER put credentials here: API keys, tokens, passwords and cloud keys belong in saved Secrets, attached with `secrets`."

addedInput schema / properties / spec / properties / secrets

Added value: +{
+  "description": "Saved secret sets this job needs, by name (e.g. [{\"name\":\"hf-training\"}]) or by the access it needs ([{\"type\":\"huggingface\",\"purpose\":\"Access a gated model\"}]). Every variable in each set is injected into the container at run time; values are never in the spec, the response or the chat. A missing set comes back with a Symbioza setup link for the user.",
+  "items": {
+    "additionalProperties": false,
+    "properties": {
+      "fields": {
+        "description": "For a custom requirement only: the environment variable NAMES the workload expects (e.g. [\"MODEL_API_TOKEN\"]). Names only, never values.",
+        "items": {
+          "pattern": "^[A-Za-z_][A-Za-z0-9_]{0,127}$",
+          "type": "string"
+        },
+        "maxItems": 32,
+        "type": "array"
+      },
+      "id": {
+        "description": "The saved set's id (sec_…). Symbioza pins it at submit; you normally give the name.",
+        "pattern": "^sec_[A-Za-z0-9_-]{22}$",
+        "type": "string"
+      },
+      "name": {
+        "description": "The saved secret set's name, e.g. \"hf-training\". Read the names with listSecrets.",
+        "pattern": "^[a-z0-9][a-z0-9_-]{0,62}$",
+        "type": "string"
+      },
+      "purpose": {
+        "description": "A short reason shown to the user on the setup page, e.g. \"Access a gated model\". Never a credential.",
+        "maxLength": 160,
+        "minLength": 1,
+        "type": "string"
+      },
+      "type": {
+        "description": "What access the workload needs: huggingface, object-storage, wandb or custom. With no name, Symbioza uses your one saved set of this type, or returns a setup link when there is none.",
+        "enum": [
+          "huggingface",
+          "object-storage",
+          "wandb",
+          "custom"
+        ],
+        "type": "string"
+      }
+    },
+    "type": "object"
+  },
+  "maxItems": 16,
+  "type": "array"
+}

addedOutput schema / properties / access

Added value: +{
+  "additionalProperties": {},
+  "description": "The saved secrets this spec attaches, by name and type only",
+  "properties": {
+    "complete": {
+      "description": "true when every set the spec names is saved and nothing collides; submitJob refuses otherwise",
+      "type": "boolean"
+    },
+    "conflicts": {
+      "description": "A variable two attached sets (or a set and env) both define",
+      "items": {
+        "type": "string"
+      },
+      "type": "array"
+    },
+    "nextAction": {
+      "anyOf": [
+        {
+          "type": "string"
+        },
+        {
+          "type": "null"
+        }
+      ],
+      "description": "add_access | fix_spec | null"
+    },
+    "ready": {
+      "items": {
+        "additionalProperties": {},
+        "properties": {
+          "fields": {
+            "items": {
+              "type": "string"
+            },
+            "type": "array"
+          },
+          "name": {
+            "type": "string"
+          },
+          "type": {
+            "type": "string"
+          }
+        },
+        "required": [
+          "name",
+          "type",
+          "fields"
+        ],
+        "type": "object"
+      },
+      "type": "array"
+    },
+    "required": {
+      "items": {
+        "additionalProperties": {},
+        "properties": {
+          "candidates": {
+            "description": "ambiguous: the saved set names that could serve; ask the user which",
+            "items": {
+              "type": "string"
+            },
+            "type": "array"
+          },
+          "message": {
+            "type": "string"
+          },
+          "name": {
+            "type": "string"
+          },
+          "purpose": {
+            "type": "string"
+          },
+          "setupUrl": {
+            "description": "missing: the Symbioza link where the user saves it; give it as returned",
+            "type": "string"
+          },
+          "status": {
+            "description": "missing | ambiguous | type_mismatch",
+            "type": "string"
+          },
+          "type": {
+            "anyOf": [
+              {
+                "type": "string"
+              },
+              {
+                "type": "null"
+              }
+            ]
+          }
+        },
+        "required": [
+          "status",
+          "type",
+          "message"
+        ],
+        "type": "object"
+      },
+      "type": "array"
+    }
+  },
+  "required": [
+    "complete",
+    "ready",
+    "required",
+    "conflicts",
+    "nextAction"
+  ],
+  "type": "object"
+}

addedOutput schema / properties / account / properties / shortfallUsd

Added value: +{
+  "description": "Whole dollars of credit this job still needs; 0 when it is covered",
+  "type": "number"
+}

changedOutput schema / properties / action / description

Previous value: -"The one next step: authenticate, fix_spec, raise_budget, add_funds, claim_credit, relax_spec_or_retry, retry_later, contact_support"New value: +"The one next step: authenticate, fix_spec, raise_budget, add_funds, claim_credit, add_access, relax_spec_or_retry, retry_later, contact_support"

changedOutput schema / properties / code / description

Previous value: -"Machine-readable refusal code: unauthorized, invalid_spec, budget_too_low, insufficient_balance, no_compute, unavailable, refused"New value: +"Machine-readable refusal code: unauthorized, invalid_spec, budget_too_low, insufficient_balance, access_required, secret_conflict, no_compute, unavailable, refused"

ChangedlistJobs2 fields changed

changedOutput schema / properties / action / description

Previous value: -"The one next step: authenticate, fix_spec, raise_budget, add_funds, claim_credit, relax_spec_or_retry, retry_later, contact_support"New value: +"The one next step: authenticate, fix_spec, raise_budget, add_funds, claim_credit, add_access, relax_spec_or_retry, retry_later, contact_support"

changedOutput schema / properties / code / description

Previous value: -"Machine-readable refusal code: unauthorized, invalid_spec, budget_too_low, insufficient_balance, no_compute, unavailable, refused"New value: +"Machine-readable refusal code: unauthorized, invalid_spec, budget_too_low, insufficient_balance, access_required, secret_conflict, no_compute, unavailable, refused"

AddedlistSecrets

ChangedsubmitJob7 fields changed

changedInput schema / properties / spec / properties / env / description

Previous value: -"Environment variables for the container. They are stored with the job spec in clear on our control plane and are included in our routine backups, so do not put credentials or other secrets here — pass them in at run time or fetch them from your own store inside the command."New value: +"Non-sensitive configuration for the container (e.g. MODEL_NAME, EPOCHS). They are stored with the job spec in clear on our control plane and are included in our routine backups, so NEVER put credentials here: API keys, tokens, passwords and cloud keys belong in saved Secrets, attached with `secrets`."

addedInput schema / properties / spec / properties / secrets

Added value: +{
+  "description": "Saved secret sets this job needs, by name (e.g. [{\"name\":\"hf-training\"}]) or by the access it needs ([{\"type\":\"huggingface\",\"purpose\":\"Access a gated model\"}]). Every variable in each set is injected into the container at run time; values are never in the spec, the response or the chat. A missing set comes back with a Symbioza setup link for the user.",
+  "items": {
+    "additionalProperties": false,
+    "properties": {
+      "fields": {
+        "description": "For a custom requirement only: the environment variable NAMES the workload expects (e.g. [\"MODEL_API_TOKEN\"]). Names only, never values.",
+        "items": {
+          "pattern": "^[A-Za-z_][A-Za-z0-9_]{0,127}$",
+          "type": "string"
+        },
+        "maxItems": 32,
+        "type": "array"
+      },
+      "id": {
+        "description": "The saved set's id (sec_…). Symbioza pins it at submit; you normally give the name.",
+        "pattern": "^sec_[A-Za-z0-9_-]{22}$",
+        "type": "string"
+      },
+      "name": {
+        "description": "The saved secret set's name, e.g. \"hf-training\". Read the names with listSecrets.",
+        "pattern": "^[a-z0-9][a-z0-9_-]{0,62}$",
+        "type": "string"
+      },
+      "purpose": {
+        "description": "A short reason shown to the user on the setup page, e.g. \"Access a gated model\". Never a credential.",
+        "maxLength": 160,
+        "minLength": 1,
+        "type": "string"
+      },
+      "type": {
+        "description": "What access the workload needs: huggingface, object-storage, wandb or custom. With no name, Symbioza uses your one saved set of this type, or returns a setup link when there is none.",
+        "enum": [
+          "huggingface",
+          "object-storage",
+          "wandb",
+          "custom"
+        ],
+        "type": "string"
+      }
+    },
+    "type": "object"
+  },
+  "maxItems": 16,
+  "type": "array"
+}

addedOutput schema / properties / access

Added value: +{
+  "additionalProperties": {},
+  "description": "On access_required or secret_conflict: what the spec's secrets need, with setup links",
+  "properties": {
+    "complete": {
+      "description": "true when every set the spec names is saved and nothing collides; submitJob refuses otherwise",
+      "type": "boolean"
+    },
+    "conflicts": {
+      "description": "A variable two attached sets (or a set and env) both define",
+      "items": {
+        "type": "string"
+      },
+      "type": "array"
+    },
+    "nextAction": {
+      "anyOf": [
+        {
+          "type": "string"
+        },
+        {
+          "type": "null"
+        }
+      ],
+      "description": "add_access | fix_spec | null"
+    },
+    "ready": {
+      "items": {
+        "additionalProperties": {},
+        "properties": {
+          "fields": {
+            "items": {
+              "type": "string"
+            },
+            "type": "array"
+          },
+          "name": {
+            "type": "string"
+          },
+          "type": {
+            "type": "string"
+          }
+        },
+        "required": [
+          "name",
+          "type",
+          "fields"
+        ],
+        "type": "object"
+      },
+      "type": "array"
+    },
+    "required": {
+      "items": {
+        "additionalProperties": {},
+        "properties": {
+          "candidates": {
+            "description": "ambiguous: the saved set names that could serve; ask the user which",
+            "items": {
+              "type": "string"
+            },
+            "type": "array"
+          },
+          "message": {
+            "type": "string"
+          },
+          "name": {
+            "type": "string"
+          },
+          "purpose": {
+            "type": "string"
+          },
+          "setupUrl": {
+            "description": "missing: the Symbioza link where the user saves it; give it as returned",
+            "type": "string"
+          },
+          "status": {
+            "description": "missing | ambiguous | type_mismatch",
+            "type": "string"
+          },
+          "type": {
+            "anyOf": [
+              {
+                "type": "string"
+              },
+              {
+                "type": "null"
+              }
+            ]
+          }
+        },
+        "required": [
+          "status",
+          "type",
+          "message"
+        ],
+        "type": "object"
+      },
+      "type": "array"
+    }
+  },
+  "required": [
+    "complete",
+    "ready",
+    "required",
+    "conflicts",
+    "nextAction"
+  ],
+  "type": "object"
+}

changedOutput schema / properties / action / description

Previous value: -"The one next step: authenticate, fix_spec, raise_budget, add_funds, claim_credit, relax_spec_or_retry, retry_later, contact_support"New value: +"The one next step: authenticate, fix_spec, raise_budget, add_funds, claim_credit, add_access, relax_spec_or_retry, retry_later, contact_support"

changedOutput schema / properties / code / description

Previous value: -"Machine-readable refusal code: unauthorized, invalid_spec, budget_too_low, insufficient_balance, no_compute, unavailable, refused"New value: +"Machine-readable refusal code: unauthorized, invalid_spec, budget_too_low, insufficient_balance, access_required, secret_conflict, no_compute, unavailable, refused"

addedOutput schema / properties / shortfallUsd

Added value: +{
+  "description": "On insufficient_balance: whole dollars of credit this job still needs",
+  "type": "number"
+}

addedOutput schema / properties / topupUrl

Added value: +{
+  "description": "On insufficient_balance: the link to give the user for that credit",
+  "type": "string"
+}

7 tool updatesSep 29, 2026
- First observedcancelJob
- First observeddescribeDataset
- First observedestimateExecution
- First observedgetArtifact
- First observedgetStatus
- First observedlistJobs
- First observedsubmitJob

Publisher details

Operator: Symbioza · Publisher source
Operator website: https://symbioza.dev · Publisher source
Vendor relationship: First-party
Documentation: https://symbioza.dev/agent · Publisher source
Trust center: Not available
Restrictions: Estimates are free; running a job spends prepaid credit, never more than the budget set for that job. · Publisher source

Related MCP Servers

Adologyofficial
Marketing Social Media
adologyai
A
license
Not graded
quality
B
maintenance
Perplexity for social: ask what competitors are running across social and get answers grounded in real ad and creative data, not guesses.
Updated a month ago (2026-09-11 19:20 UTC)
2
Apache 2.0
Competitor Monitor AI MCP
Marketing Research & Data AI & Machine Learning
CSOAI-ORG
A
license
C
quality
B
maintenance
Competitor Monitor AI - MCP server providing AI-powered tools and automation by MEOK AI Labs
Updated 12 days ago (2026-10-08 01:01 UTC)
11
9 npm
MIT
Citesurf MCP
Marketing Search Monitoring
citesurf
A
license
A
quality
C
maintenance
Enables brand visibility monitoring across major AI platforms like ChatGPT, Claude, Gemini, and Perplexity. It allows users to track visibility scores, analyze competitor data, and receive actionable insights to improve AI-generated brand recommendations.
Updated 6 months ago (2026-04-23 11:46 UTC)
16
11 npm
1
MIT
industrylens-mcpofficial
Research & Data
IndustryLensOp
A
license
Not graded
quality
B
maintenance
Browse IndustryLens's published competitive-intelligence reports and head-to-head competitor comparisons from any AI agent — real, source-backed data.
Updated 2 months ago (2026-08-17 15:55 UTC)
MIT