The iai-mcp server provides a local, private, and encrypted long-term memory system for AI assistants (primarily Claude), enabling persistent capture, recall, and consolidation of conversation history across sessions. All data stays on your machine (AES-256-GCM encrypted, Unix socket, no telemetry).
Core Memory Operations
memory_recall — Retrieve verbatim memories via semantic similarity, graph-link strength, and recency
memory_recall_structural — Structural recall using role-filler bindings (e.g., {"agent": "Alice"})
memory_capture — Store a turn (user/assistant/system) into a memory tier (episodic, semantic, procedural, parametric, or working), with auto-deduplication at cosine ≥ 0.95
memory_reinforce — Strengthen Hebbian edges between co-retrieved records
memory_contradict — Mark a record as outdated and store an updated fact
memory_consolidate — Trigger background merging of duplicates, edge decay, and semantic summary building
Profile & Personalization
profile_get_set — Read/write 11 sealed profile knobs (AUTIST parameters + wake_depth) capturing learned user preferences and style cues
Introspection & Analytics
topology — Memory graph snapshot (node/edge count, clustering, community count)
schema_list — List induced patterns/schemas, filterable by domain and confidence
events_query — Query system events by kind, severity, and time range
curiosity_pending — List pending curiosity questions generated by the system
camouflaging_status — Camouflaging detection status over a configurable weekly window
Design targets: ≥99% verbatim recall at 10k items, <100ms p95 latency. Primarily tested on macOS (Apple Silicon) with Python 3.11/3.12 and Node.js 18+.
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@iai-personal-memory-engine what was my last session about?"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
English | 中文
iai-memory gives the coding assistant you already use a persistent memory on your machine. With ambient hooks enabled, it records both sides of a conversation, keeps the captured wording, and supplies a bounded slice of relevant history when a session starts or advances. You do not maintain a memory file or keep saying “remember this.”
Corrections do not rewrite history. A changed fact becomes a new record linked to the superseded one, so both the current statement and the earlier wording remain queryable. Recall can return contradictory or superseded records beside matching ones instead of letting an obsolete fact pass as current.
This is a personal engine for an assistant you already use, not a multi-tenant memory API for an application. Episodic capture is write-once and verbatim; storage, embeddings, retrieval, graph operations, and the dashboard run locally. No external vector or graph database is required.
The memory style is autistic by design: verbatim over paraphrase, precise cues, sustained focus, and rare events kept rare. Why the name.
Related MCP server: local-agent-context
python3.12 -m pip install -U iai-pmeThen run inside Claude Code:
/plugin marketplace add CodeAbra/iai-personal-memory-engine
/plugin install iai-memory@iai-pmeRestart the session, then verify:
iai --version
iai-mcp daemon status
iai-mcp doctorPython 3.11 is also supported.
curl -fsSL https://raw.githubusercontent.com/CodeAbra/iai-personal-memory-engine/main/scripts/bootstrap.sh | bashThis builds the Rust engine and TypeScript wrapper, installs the background service and hooks, registers Claude Code, and runs the health check. It requires Git, Python 3.11/3.12, Node.js 18+, and Rust. To inspect the steps without changing anything:
curl -fsSL https://raw.githubusercontent.com/CodeAbra/iai-personal-memory-engine/main/scripts/bootstrap.sh | bash -s -- --dry-runpython3.12 -m pip install -U iai-pme
iai-mcp crypto init
iai-mcp daemon install
iai-mcp capture-hooks install --target codexReplace codex with cursor, antigravity, hermes, openclaw, or all.
MCP tools work with any MCP-over-stdio client; automatic capture and context
injection depend on the hooks exposed by the host. See the
technical reference.
New stores use the native engine format by default; an existing store keeps its
current format on upgrade. To move an existing legacy SQLite store onto the
native engine, run iai-mcp migrate-to-lilli — iai-mcp doctor prints the exact
command, and the technical reference documents the full flow.
Event | Action |
Prompt | New turns are appended to a session buffer as file IO; no embedding or engine RPC is needed on the capture path |
Session end | Remaining transcript content is rolled over for ingestion; hook failures do not block the host |
Session start | A bounded memory prefix is exposed as host context; an empty store or unavailable engine yields empty output |
Later turns | Supported hosts receive a small foresight or delta pack with age and revision markers |
Idle time | Captures are embedded, deduplicated, encrypted, inserted, clustered, consolidated, reinforced, and decayed |
The background process is called the daemon in the CLI. The MCP wrapper and
iai can still read the local store directly when it is asleep or temporarily
unavailable.
Tier | Contains |
Episodic | Timestamped, write-once fragments of what was said |
Semantic | Summaries induced from related episodes during idle consolidation |
Procedural | Ten bounded behavioural parameters learned over time |
Distinct hyperdimensional representations keep literal detail, semantic structure, and behavioural tendencies from collapsing into one vector surface.
The local, LLM-free recall path combines semantic similarity, graph evidence,
recency, temporal validity, and lexical evidence. memory_recall returns both
hits and anti_hits; memory_contradict closes the old record's validity
interval, creates a new record, and links the two.
While idle, the engine groups related episodes, induces semantic memory,
reinforces useful paths, and decays weak unreviewed edges. One optional REM step
may invoke claude -p through the user's existing Claude subscription, capped
at no more than 1% of the daily quota. No Anthropic API key is required.
Component | Role |
Hippo | Encrypted records, vector index, and graph in one local store |
MOSAIC | Leiden-family community detection with stable community identity |
Lilli HD | Hyperdimensional substrate and structural recall |
Native engine | Rust embedder and graph kernels |
iai brainThe local dashboard searches the store, exposes graph neighbourhoods and contradictions, pins or fades memories, ingests files, controls the background engine, and reports token-use estimates from your own store.
iai recall · temporal-recall · search · ask · capture · teach · upload
iai watch · brain · status · lastiai upload accepts documents, Office files, e-books, source code,
configuration files, and directories. Full formats and administrative commands
are listed in docs/REFERENCE.md.
Every harness ships in bench/; methodology and reproduce commands are in
BENCHMARKS.md.
Benchmark | Result |
Rescue@10 after contradiction | 1.000 |
Historical-verbatim hit@10 | 1.000 |
LongMemEval-S R@5, product embedder | 0.962 |
LongMemEval-S R@10, product embedder | 0.978 |
Historical-verbatim retrieval uses a flat-cosine baseline of about 0.71. With
the matched all-MiniLM-L6-v2 embedder, iai-memory and mempalace v3.3.6 both
score R@5 0.966 and R@10 0.978; no win is claimed.
On the author's store, an automatically injected memory pack averaged about
350 tokens versus about 2,850 tokens for the agent-search round trip it
replaced: approximately 88% cheaper on that measured workload. This does not
apply to explicit memory_recall, whose default response budget is 1,500
tokens.
memory_recall memory_temporal_recall
memory_recall_structural memory_search
memory_capture memory_contradict
memory_reinforce memory_consolidate
profile_get_set topology
schema_list events_query
episodes_recent curiosity_pendingFourteen tools cover cue, temporal, structural, and lexical recall; capture and correction; reinforcement and consolidation; behavioural-profile control; and store introspection.
Host | Ambient behaviour |
Claude Code | Session-start recall, per-turn updates, turn capture, and session capture |
Codex CLI | Full integration through Codex hooks |
Cursor | Session-start recall and capture; no per-turn text injection |
Antigravity | Recall per invocation and lossless transcript capture |
Hermes 0.5.0+ | Recall before model calls and capture from its message store |
OpenClaw | MCP tools on request; no ambient shell hooks |
Gemini CLI and other MCP hosts | MCP tools; no bundled host-specific hooks unless listed above |
Claude Desktop | MCP tools; plain Chat does not expose Claude Code-style ambient hooks |
Records are encrypted at rest with AES-256-GCM. The store and key live under
~/.iai-mcp/; back them up together.
macOS and Linux use a Unix socket. Windows uses an ephemeral loopback port with a per-user token.
There is no iai-memory account, telemetry pipeline, hosted dashboard, or cross-machine sync.
Optional iai-memory network activity is the REM claude -p step and a daily
PyPI version check. Set IAI_MCP_VERSION_CHECK=0 to disable the check.
The store refuses to mix incompatible embedding generations; changing the embedder requires an explicit migration.
Recall is usually mediocre during roughly the first ten sessions, and quality and latency depend on corpus size, language, embedder, and stored history.
The default store is English-first. Raw non-English records require an
explicit raw:<lang> tag and a multilingual or custom embedder.
Windows support is beta. Ambient behaviour varies with host hook support.
The project is solo-maintained and has no enterprise SLA.
Health and updates:
iai-mcp doctor # 36 checks
iai-mcp daemon status
iai-mcp self-updateIAI — Independent Autistic Intelligence describes the memory design.
Independent: the engine, store, embeddings, and dashboard run locally.
Autistic: literal preservation, precise cues, sustained focus, and rare events retained as rare rather than smoothed into a typical summary. This is an operational design description, not a diagnosis or casual metaphor.
Intelligence: used in the systems sense — a process that observes, adapts, reorganizes itself, and remains viable over time.
“Personal memory engine” describes the scope: one person's memory, on one machine, used by the assistant they already have.
docs/REFERENCE.md — technical and operational reference
BENCHMARKS.md — methodology and reproduce commands
docs/EMBEDDERS.md — providers, languages, and migrations
CHANGELOG.md — release history
CONTRIBUTING.md — development and test setup
SECURITY.md — private vulnerability reporting
Issues and pull requests are welcome. Changes to retrieval, capture, contradiction handling, or consolidation should include relevant benchmark reruns.
By Areg Aramovich Noya and Lilli Noya, in collaboration with the team at lcgc.dev.
Check a claim (e.g. 'X is not done') against memory. Returns hits + anti_hits + a freshness verdict in one call.
| Name | Required | Description | Default |
|---|---|---|---|
| cue | Yes | The claim to check, e.g. 'the dashboard is not built yet'. | |
| session_id | No | Current session id; gets written into provenance. Omit to use '-'. | |
| budget_tokens | No | Soft token budget for the underlying recall (default 1500). |
| Name | Required | Description |
|---|---|---|
| hits | No | |
| _source | No | |
| verdict | No | |
| anti_hits | No | |
| verdict_reason | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already signal read-only, idempotent, non-destructive behavior. The description adds value by disclosing that the call returns both supporting (hits) and contradicting (anti_hits) evidence plus a freshness verdict, and that this is combined into one call. It does not discuss failure modes or edge cases, but the output schema covers the return shape and the operation is simple.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
One sentence packs the action, the example, the result summary, and the one-call characteristic with no filler. The key behavior is front-loaded before the details.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a simple three-parameter tool with full schema descriptions, read-only/idempotent annotations, and an output schema, the description is sufficient: it tells the agent what claim checking means and what to expect. Nothing critical for calling the tool correctly is missing.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100%, so the schema already documents all three parameters. The description adds no parameter-level meaning beyond labeling the primary input as a claim, and the schema already gives an example for cue. Baseline 3 is appropriate.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
Description names a specific verb ('Check') and resource ('claim ... against memory'), and the expected result ('hits + anti_hits + a freshness verdict in one call') makes it distinguishable from raw memory search/recall siblings. It goes beyond the tool name to define what a claim check actually produces.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The purpose statement implies when to use this tool: when an agent has a claim to verify and wants evidence plus a freshness assessment in a single call. However, it does not name sibling alternatives (e.g., memory_search, memory_recall) or state when to prefer those instead, so the routing guidance is only implicit.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
List pending curiosity questions queued by the sleep daemon. Read-only. Filter by session_id.
| Name | Required | Description | Default |
|---|---|---|---|
| session_id | No | Only return questions from this session. Omit to return questions from every session in the queue. |
| Name | Required | Description |
|---|---|---|
| count | No | |
| questions | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnlyHint=true, idempotentHint=true, and destructiveHint=false. The description adds context that the questions are 'queued by the sleep daemon' and are 'pending,' which is extra context beyond the annotations, but not extensive. It does not contradict annotations.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is extremely concise: three short sentences, front-loaded with the main action, then read-only note, then filter instruction. Every word earns its place with no redundancy or fluff.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a simple list tool with one optional parameter, a rich set of annotations, and an output schema, the description is complete. It specifies the source (sleep daemon), state (pending), and filtering option, covering all necessary context.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100% (session_id has a clear description). The description mentions 'Filter by session_id' but does not add meaning beyond the schema. Baseline of 3 applies since the schema already documents the parameter well.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states a specific action and resource: 'List pending curiosity questions queued by the sleep daemon.' It also notes the tool is read-only and can be filtered by session_id, distinguishing it from sibling tools focused on memory operations.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Provides clear context on what the tool does (lists pending curiosity questions) and how to narrow results (filter by session_id). It lacks explicit alternatives or when-not-to-use guidance, but the read-only nature and filtering hint are useful.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Returns the N most-recent user-turn records, time-desc. Optional session_id filter. GLOBAL across all projects.
| Name | Required | Description | Default |
|---|---|---|---|
| n | No | How many turns to return (default 10, max 1000). | |
| session_id | No | Filter to a specific session UUID. |
| Name | Required | Description |
|---|---|---|
| count | No | |
| turns | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already cover read-only and idempotent behavior. The description adds valuable context beyond that: the "GLOBAL across all projects" scope, which could be surprising for users expecting project isolation, and the time-desc ordering. However, it does not disclose potential rate limits, pagination, or what happens when n exceeds the max, though those are partially covered by the schema.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single, well-structured sentence that front-loads the core functionality, then adds the optional filter and global scope. Every clause adds value with no redundancy or filler.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a simple read-only list tool with two optional parameters and a rich output schema, the description covers the essential behavior, scope, and ordering. The annotations and schema fill in safety and parameter details, making this sufficiently complete for a correct invocation.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100% (n and session_id are fully described). The description only reiterates "Optional session_id filter," which adds no new meaning beyond the schema. It does not clarify edge cases like n=0 or negative values, but the schema already provides defaults and bounds.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the verb "Returns" with a specific resource: "N most-recent user-turn records," including ordering (time-desc) and an optional filter. It also distinguishes itself by declaring "GLOBAL across all projects," setting it apart from project-scoped siblings.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description implicitly communicates when to use it (for recent user-turn records globally) but provides no explicit guidance on when not to use it or what alternatives might be better (e.g., memory_search for semantic recall, events_query for event logs). It lacks explicit exclusions or alternative tool references.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Query user-visible events (kind whitelist). Read-only. Optional since (ISO-8601), severity, limit.
| Name | Required | Description | Default |
|---|---|---|---|
| kind | Yes | Event kind. Must be in the whitelist (s4_contradiction, trajectory_metric, ...). | |
| limit | No | Maximum events returned (default 100, capped at 1000 by the daemon regardless of the value supplied). | |
| since | No | ISO-8601 timestamp; only events at or after this are returned. Omit to return events from the start of the log. | |
| severity | No | Optional severity filter. Omit to return all severities. |
| Name | Required | Description |
|---|---|---|
| count | No | |
| events | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnly, idempotent, and non-destructive. The description adds context about 'user-visible events' and the 'kind whitelist', which are meaningful behavioral constraints beyond the annotations. No contradiction.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Single concise sentence that front-loads the primary purpose and lists key filters. Every word earns its place; no redundancy.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
With full schema, annotations, and an output schema present, the description is sufficiently complete for a read-only query tool. It communicates the essential scope and filters; return format is handled by output schema.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100%, so baseline is 3. The description merely lists the optional parameters without adding details beyond what the schema already provides (e.g., ISO-8601 for since, enums for severity). No extra semantic value.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description uses a specific verb 'Query' with a clear resource 'user-visible events' and scope ('kind whitelist'). This distinguishes it from sibling tools like memory_recall or episodes_recent.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description clearly indicates this is the tool for querying events, with optional filters (since, severity, limit) and a whitelist constraint. It does not explicitly compare to alternatives, but the context is clear and there are no exclusions mentioned.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Capture a verbatim turn (auto-dedups near-duplicates). Use for corrections, not for minting standing-order directives.
| Name | Required | Description | Default |
|---|---|---|---|
| cue | No | Short natural-language cue used for embedding + dedup lookup. If empty, `text` itself is embedded. | |
| role | No | Who produced this turn — tags the record for filtering. | user |
| text | Yes | Verbatim text to capture (user utterance, Claude decision, or observation). Min 12 chars, max 8000 (longer is truncated). | |
| tier | No | Memory tier. Default 'episodic' (verbatim user utterances). Use 'semantic' for induced summaries, 'procedural' for learned behaviour notes. | episodic |
| focus | No | Optional current point of attention for the live session task. Folded verbatim onto this session's own working-tier entry after the capture completes, alongside next_action. | |
| agent_id | No | Optional id of a background agent this capture is spawning or completing. Combine with agent_role and agent_expected_artifact to register a pending agent; combine with agent_complete_id on a later call to mark it done. | |
| agent_role | No | Optional role of the spawned background agent (for example 'research' or 'implement'). Required alongside agent_id and agent_expected_artifact to register a spawn; omitted otherwise. | |
| session_id | No | Current session id for provenance. | |
| agent_model | No | Optional model label for the spawned background agent, recorded on the registry entry when agent_id/agent_role/agent_expected_artifact register a spawn. | |
| next_action | No | Optional immediate next step for the current live session task. Folded verbatim onto this session's own working-tier entry after the capture completes; surfaces at the next session start and on every subsequent turn until updated again. | |
| salience_level | No | Caller-declared salience level for a decision, correction, or load-bearing preference marked in-turn. Additive rank-fusion boost only -- never a merge/drop lock. Omit for 'unflagged' (default, no behavior change). A value outside the enum is coerced to 'unflagged' server-side, never rejected. | unflagged |
| epistemic_status | No | Caller-declared epistemic status. Omit for 'unknown' (default, no behavior change). A value outside the enum is coerced to 'unknown' server-side, never rejected. | unknown |
| agent_complete_id | No | Optional id of a previously spawned background agent to mark complete on this call. | |
| agent_expected_artifact | No | Optional artifact the spawned background agent is expected to produce. Required alongside agent_id and agent_role to register a spawn; omitted otherwise. |
| Name | Required | Description |
|---|---|---|
| reason | No | |
| status | No | |
| record_id | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The description discloses one useful behavioral trait beyond the annotations: 'auto-dedups near-duplicates.' However, it does not explain side effects, persistence, merge/drop behavior, or response semantics. The annotations are all false, so they provide no additional safety or behavior hints, leaving the description with more responsibility than it fully carries.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single, front-loaded sentence with no filler. It states the core behavior, a key behavioral detail, and usage guidance, all in under 20 words.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a tool with 14 parameters and several conditional agent-spawning fields, the text description is brief. The rich 100%-coverage schema and presence of an output schema compensate substantially, but the description itself does not hint at the background-agent registration/completion capabilities, so it is not fully complete.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100%, so all 14 parameters already have detailed descriptions. The tool description adds no new parameter-level meaning, and with full schema coverage the baseline of 3 is appropriate.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
Description states a specific action ('Capture a verbatim turn') and a precise resource/scope ('verbatim turn'), and it even carves out what it is not for ('not for minting standing-order directives'). This clearly distinguishes it from sibling memory tools like memory_recall or memory_consolidate.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description gives explicit when-to-use guidance ('Use for corrections') and an explicit exclusion ('not for minting standing-order directives'). It stops short of naming an alternative tool to use for directives, so it lacks the full when/when-not/alternatives triad.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Trigger sleep-cycle consolidation: schema induction, FSRS decay, Hebbian pruning. Mutates store; idempotent in one sleep window.
| Name | Required | Description | Default |
|---|---|---|---|
| session_id | No | Optional session id used for provenance tagging on the consolidate event. Defaults to '-' when omitted. |
| Name | Required | Description |
|---|---|---|
| mode | No | |
| tier | No | |
| decay_result | No | |
| schema_candidates | No | |
| summaries_created | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Description adds value beyond annotations by stating 'Mutates store' (aligning with readOnlyHint: false) and specifying 'idempotent in one sleep window' (extending the idempotentHint). It also discloses the internal processes involved. No contradiction with annotations.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Two concise sentences with essential information front-loaded. Every word earns its place, with no redundancy or filler.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a tool with one optional parameter, an output schema, and annotations covering safety/idempotency, the description provides sufficient context about purpose and effects. It could include more about when consolidation is appropriate, but the current level is adequate for selection and invocation.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100% and the single parameter (session_id) has a complete description in the schema. The tool description adds no additional semantic meaning about the parameter, so baseline 3 is appropriate.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
Description clearly states the tool's function: 'Trigger sleep-cycle consolidation' and specifies three distinct processes (schema induction, FSRS decay, Hebbian pruning). This specific verb+resource combination distinguishes it from sibling tools like memory_recall or memory_reinforce.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description implies when to use this tool (when triggering sleep-cycle consolidation) but does not explicitly state when not to use it or mention alternatives. Clear context but no exclusions or comparisons to sibling tools.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Mark a record contradicted; new fact stored as a NEW record (old NEVER deleted). Mutates store.
| Name | Required | Description | Default |
|---|---|---|---|
| id | Yes | UUID of the record being contradicted. | |
| new_fact | Yes | The updated verbatim fact. Stored as a new record; the old record is preserved (episodic write-once) and linked via a `contradicts` edge. | |
| cue_embedding | No | Optional pre-computed embedding vector for the contradicting fact (EMBED_DIM=384 floats; bge-small-en-v1.5). When omitted, the daemon embeds new_fact server-side. | |
| epistemic_status | No | Caller-declared epistemic status of the corrected fact. Omit for 'unknown' (default, no behavior change). A value outside the enum is coerced to 'unknown' server-side, never rejected. | unknown |
| Name | Required | Description |
|---|---|---|
| ts | No | |
| edge_type | No | |
| original_id | No | |
| new_record_id | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already show this is a mutating, non-destructive operation. The description adds valuable behavioral clarity by explicitly stating that the old record is NEVER deleted and that the store is mutated. This goes beyond the raw annotation flags and sets correct expectations about data preservation.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single front-loaded sentence that immediately states the action, the key side effect, and the safety guarantee. Every clause earns its place; there is no redundant or vague filler.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the rich input schema, annotations, and presence of an output schema, this description is sufficient. It covers the essential behavioral contract—mutation, preservation of old records, and creation of a new contradicted record—without needing to restate schema details.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100%, and each parameter including id, new_fact, cue_embedding, and epistemic_status already has a detailed schema description. The tool description contributes no additional parameter-level nuance, so the baseline of 3 is appropriate.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description uses a specific verb and resource: 'Mark a record contradicted' and clearly states the core effect—new fact stored as a NEW record while the old is never deleted. This differentiates the tool from siblings like memory_capture or memory_reinforce without needing to name them.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description implies the usage—contradicting a record while preserving the old fact—but gives no explicit guidance about when to prefer this over memory_capture, memory_reinforce, or other alternatives. There are no stated exclusions or conditional routing cues.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recall verbatim memories by cue — decisions, preferences, prior discussion, rationale. Call before a repository search. Returns hits + anti_hits.
| Name | Required | Description | Default |
|---|---|---|---|
| cue | Yes | Natural-language query to match against stored memories. Embedded server-side via bge-small-en-v1.5 (384d) unless `cue_embedding` is supplied. | |
| language | No | Optional ISO-639-1 language hint for the sleep-suggestion path (8 supported: en/ru/ja/ar/de/fr/es/zh). Defaults to 'en' when omitted. Hot-path retrieval is language-agnostic; this key only affects the sleep-suggestion regex pre-screen. | |
| session_id | No | Current session id; gets written into every recalled record's provenance. Omit to use '-'. | |
| budget_tokens | No | Soft token budget for the response (default 1500). Hits are appended until the next would exceed this budget; at least one hit is always returned. | |
| cue_embedding | No | Optional pre-computed embedding vector for the cue (EMBED_DIM=384 floats; bge-small-en-v1.5). When omitted, the daemon embeds the cue server-side. Used by memory_contradict and tests that need byte-stable embeddings. |
| Name | Required | Description |
|---|---|---|
| hits | No | |
| hints | No | |
| cue_mode | No | |
| anti_hits | No | |
| budget_used | No | |
| ann_path_used | No | |
| pask_teachback | No | |
| activation_trace | No | |
| overnight_digest | No | |
| patterns_observed | No | |
| exact_authority_used | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnlyHint, idempotentHint, and destructiveHint. The description adds that recall is 'verbatim' and that it 'Returns hits + anti_hits,' providing behavioral expectations beyond the annotations. It does not contradict annotations and adds context.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is only two sentences, front-loaded with the action ('Recall'), and each clause carries distinct information: what it recalls, when to call it, and what it returns. No filler or redundancy.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the complete schema, output schema, and rich annotations, the description covers purpose, usage timing, and a return-shape hint. It could mention anti_hits semantics or caveats, but the schema and output schema fill those gaps. It is sufficient for a tool of this complexity.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
All 5 parameters are fully described in the schema (100% coverage), so the baseline is 3. The description identifies the 'cue' as the matching mechanism and mentions return behavior, but provides no additional parameter semantics beyond what the schema already offers.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description uses a specific verb ('Recall') and resource ('memories by cue') and lists content types ('decisions, preferences, prior discussion, rationale'). It distinguishes itself from siblings via 'verbatim' and the 'Call before a repository search' directive, though it doesn't explicitly name alternative tools.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
It gives a clear context: 'Call before a repository search.' This tells when to use it but does not explicitly state when not to use it or name alternatives. Per rubric, that's a 4 (clear context, no exclusions).
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Structural recall via TEM role->filler bindings (BSC hypervectors). Read-only. Prefer over memory_recall for role-filler queries.
| Name | Required | Description | Default |
|---|---|---|---|
| max_records | No | Hard cap on records scanned after fetch (default 5000, max 50000). Prevents accidental full-corpus scans from `{}`. | |
| budget_tokens | No | Soft token budget for the response (default 2000). Hits are appended until the next would exceed this budget. | |
| structure_query | No | Optional role->filler map, e.g. {"agent": "agent_name"}. Each value is hashed to a filler hypervector. When omitted or empty, query HV is zero-filled and every row with structure_hv is scored (expensive at large N). |
| Name | Required | Description |
|---|---|---|
| hits | No | |
| anti_hits | No | |
| budget_used | No | |
| activation_trace | No | |
| structural_query_size | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnlyHint=true, idempotentHint=true, and destructiveHint=false, covering the safety profile. The description adds minimal behavioral context beyond the annotations, such as the TEM role->filler binding mechanism and the 'Read-only' statement, but these do not significantly enrich the agent's understanding of operational behavior. It does not contradict the annotations, though it also does not disclose return format or side effects beyond what annotations already imply.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is extremely concise, consisting of three short phrases that convey purpose, safety, and usage preference. Every word adds value, and it is front-loaded with the core concept. No unnecessary filler or repetition exists.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
The tool has a detailed input schema with all parameters documented, a rich output schema, and strong annotations. The description covers the core purpose and usage differentiation. It does not explain return values, but the output schema presumably handles that, and the overall definition is comprehensive enough for an agent to select and invoke the tool correctly.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100%, and the parameter descriptions are detailed, including hard caps, budget behavior, and the meaning of omitted structure_query. The description text itself does not add further parameter semantics beyond relating the tool to 'role-filler queries', which is already evident from the schema. Therefore the baseline score of 3 is appropriate.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool performs structural recall via TEM role->filler bindings (BSC hypervectors), giving a specific mechanism and resource. It also explicitly distinguishes itself from the sibling tool memory_recall by noting it is preferred for role-filler queries, making its purpose unambiguous.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description provides explicit usage guidance: 'Prefer over memory_recall for role-filler queries' directly tells when to use this tool over an alternative. The parameter schema further explains when structure_query is omitted, warning of an expensive full-corpus scan, which supplements the when-to-use context.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Boost Hebbian edges among co-retrieved record ids. Mutates edge weights. Use when two records co-answered.
| Name | Required | Description | Default |
|---|---|---|---|
| ids | Yes | Record UUIDs that were co-retrieved in the current context. Edges between every pair are incremented; identical pair sets are idempotent within one session. | |
| session_id | No | Session identifier for correlating this reinforcement with the session's retrieval history. Optional; omit for old clients. |
| Name | Required | Description |
|---|---|---|
| new_weights | No | |
| edges_boosted | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The description discloses the mutation semantics ('Mutates edge weights') and reinforces the Hebbian reinforcement idea, which adds detail beyond the readOnlyHint=false annotation. It does not contradict the annotations; idempotency is already declared in the schema and annotation.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Three short, meaningful sentences: the action, the mutation effect, and the usage condition. No fluff or repetition that wastes tokens.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a simple two-parameter tool with an output schema and annotations, the description plus schema provide enough to invoke it correctly. It could be more explicit about when not to use it relative to memory_recall or memory_consolidate, but this is not a critical gap.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
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 only contextual meaning to 'ids' via 'co-retrieved record ids' but does not add parameter-level details that the schema already lacks.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description uses a specific verb ('Boost') with a clear resource ('Hebbian edges among co-retrieved record ids') and states the core effect ('Mutates edge weights'). It distinguishes this from sibling memory tools by focusing on edge-weight reinforcement rather than search, capture, or recall.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
It explicitly says 'Use when two records co-answered', giving a clear trigger condition. It does not name exclusions or alternative sibling tools, so it does not fully meet the 5-level bar.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Use for code/doc search; returns hints to verify — never replaces a repository search.
| Name | Required | Description | Default |
|---|---|---|---|
| k | No | Max hits (default 8, max 24). | |
| query | Yes | Search text: identifiers, phrases, or a question. |
| Name | Required | Description |
|---|---|---|
| hits | No | |
| frame | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnlyHint, idempotentHint, and non-destructive behavior. The description adds valuable context by stating that results are 'hints to verify' and that the tool does not substitute for a full repository search. This goes beyond the annotations and clarifies the reliability of the output.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single, front-loaded sentence that efficiently conveys purpose, usage, and a behavioral caveat. Every clause adds value: 'Use for code/doc search' (purpose), 'returns hints to verify' (behavior), and 'never replaces a repository search' (limitation). No wasted words.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the presence of 2 well-documented parameters, annotations, and an output schema, the description covers the essential aspects: what it does, when to use it, and a key limitation. It does not explain return values, but the output schema handles that. Minor gap: no explicit mention of how this differs from memory_recall siblings, but overall the description is sufficient for an agent to select and invoke the tool.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100%, so the schema already documents both parameters (query and k) with descriptions. The description does not add additional meaning specific to the parameters; it only mentions 'code/doc search' which indirectly implies the query type, but that is marginal. Baseline 3 is appropriate.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description states a clear verb and resource: 'Use for code/doc search'. It distinguishes itself from a repository search by saying it 'never replaces a repository search', but does not explicitly differentiate from sibling memory tools like memory_recall or memory_temporal_recall. Still, the core purpose is clear.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description provides explicit when-to-use context ('Use for code/doc search') and a when-not-to-use ('never replaces a repository search'). However, it does not mention alternatives or explicitly contrast with sibling tools, so it falls short of the full 'explicit alternatives' bar.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Time-travel recall: as_of bounds records, changed_since filters events. Read-only.
| Name | Required | Description | Default |
|---|---|---|---|
| cue | No | Optional natural-language cue. If omitted, the records side returns recency-ordered rows bounded by as_of. | |
| as_of | No | ISO-8601 timestamp. Bounds the records side: records.created_at <= as_of. | |
| limit | No | Maximum items per side (default 10). | |
| changed_since | No | ISO-8601 timestamp. Bounds the events side: events.ts > changed_since (strict). |
| Name | Required | Description |
|---|---|---|
| hits | No | |
| _scope | No | |
| changed_since_events | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnlyHint, idempotentHint, and destructiveHint, so the 'Read-only' phrase is redundant. However, the description adds valuable behavioral detail by explaining that as_of bounds the records side and changed_since filters the events side, clarifying the dual-sided temporal semantics beyond the annotations.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single, efficient sentence with a colon introducing the key mechanics. Every phrase carries meaning, and the front-loaded 'Time-travel recall' immediately conveys the tool's niche. No wasted words.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the output schema exists, return value details are not needed. The description covers the core behavior (the two bounds) and mentions the read-only nature, which aligns with annotations. It omits discussion of cue and limit, but the schema provides those. Overall, it is complete for a read-only temporal query tool.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The schema already covers all parameters (100% coverage), giving a baseline of 3. The description adds meaning by mapping as_of to the records side and changed_since to the events side, helping the agent understand how these parameters relate to the two outputs. This semantic grouping exceeds the schema's individual descriptions.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool's purpose with a specific verb ('recall') and resource ('records' and 'events') while uniquely specifying temporal scoping via 'as_of bounds records, changed_since filters events'. This distinguishes it from siblings like memory_recall and memory_search by emphasizing its temporal nature.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description implies temporal use cases but does not explicitly state when to use this tool over siblings like memory_recall or events_query. No exclusions or alternative recommendations are provided, leaving the agent to infer usage context from the temporal keywords.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Read or write a profile knob (10 sealed: 9 AUTIST + wake_depth). operation get|set; returns knob value.
| Name | Required | Description | Default |
|---|---|---|---|
| knob | No | Knob name. Omit on 'get' to retrieve all live + deferred knobs. Required on 'set'. | |
| value | No | New value when operation='set'. Any JSON-serialisable type matching the knob's declared type in the sealed registry. | |
| operation | Yes | Whether to read or write a knob. 'get' with no `knob` returns all live + deferred knob values; 'set' requires both `knob` and `value`. |
| Name | Required | Description |
|---|---|---|
No output parameters | ||
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The annotations already disclose readOnly=false, destructive=false, and idempotent=true; the description adds that the knobs are sealed and that get/set returns a value. It does not describe set-side effects such as validation, persistence, or deferred knob behavior, but the annotations carry the core safety profile.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single sentence plus a compact clause, with each component earning its place: purpose, knob inventory, operation mode, and return behavior. There is no filler or repetition of schema content.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a tool with 3 parameters, an output schema, and complete schema coverage, this description is sufficient: it states the resource, operation, knob list, and return. It could mention error cases or dynamic/deferred knob behavior, but the schema and annotations already cover much of the operational context.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100% and the schema already documents operation, knob, and value with rich details, including 'get' with no knob returns all live+deferred values and set requires both fields. The tool description adds no new parameter detail, so the baseline score is appropriate.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool 'Read or write a profile knob' and names the two operations, get and set. It further specifies the exact sealed knob set (9 AUTIST + wake_depth), which makes the purpose concrete and distinguishes it from the unrelated sibling tools.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
It gives clear context that the tool is used to access/update profile knobs and that 'operation' selects read vs write. It doesn't explicitly name alternatives or when-not-to-use cases, but the tool's domain is narrow enough that no sibling overlap is apparent.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
List induced schemas (Tier-0 + Tier-1) from sleep consolidation. Read-only. Filter by domain and confidence_min.
| Name | Required | Description | Default |
|---|---|---|---|
| domain | No | Only return schemas tagged with this domain (e.g. 'coding'). Omit to return schemas across all domains. | |
| confidence_min | No | Minimum parsed confidence (0.0-1.0). Default 0.0 returns all schemas; raise to 0.5+ to filter out low-evidence candidates. |
| Name | Required | Description |
|---|---|---|
| total | No | |
| schemas | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnlyHint, idempotentHint, and destructiveHint=false, so the safety profile is covered. The description adds useful context about the source (sleep consolidation) and the Tier-0 + Tier-1 restriction, but it does not disclose any further behavioral traits such as return format or side effects. This aligns with the calibration example where annotations carry the main burden and the description adds moderate value.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is extremely concise, with the core action and scope front-loaded in the first sentence. The second sentence adds the read-only note and filter parameters with no wasted words.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the tool's simplicity (two optional parameters, full schema descriptions, an output schema, and rich annotations), the description adequately covers the essential aspects: what it lists, the source, safety, and filtering. No significant information is missing for an agent to select and invoke the tool correctly.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The input schema already provides 100% coverage for both parameters, including defaults and semantics. The description only echoes 'Filter by domain and confidence_min' without adding any additional meaning beyond what the schema provides, so the baseline 3 is appropriate.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description uses a specific verb 'List' with a clear resource 'induced schemas' and narrows scope to 'Tier-0 + Tier-1 from sleep consolidation.' This not only states what it does but also distinguishes it from sibling memory tools like memory_recall or memory_search.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description clearly indicates the tool is read-only and mentions the two filter dimensions (domain and confidence_min), which provides clear context for when to use it. However, it does not explicitly name alternatives or exclusion criteria compared to sibling tools, so it stops short of a 5.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Snapshot of memory-graph topology: N, C, L, sigma, community_count, regime. Read-only diagnostic; sigma never toggles retrieval.
| Name | Required | Description | Default |
|---|---|---|---|
No parameters | |||
| Name | Required | Description |
|---|---|---|
| C | No | |
| L | No | |
| N | No | |
| sigma | No | |
| regime | No | |
| community_count | No | |
| rich_club_ratio | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnlyHint, idempotentHint, and destructiveHint. The description adds a specific behavioral guarantee that 'sigma never toggles retrieval', which is useful context beyond the annotations. It also clarifies the read-only nature without contradicting annotations.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single, information-dense sentence. It front-loads the core purpose, lists the key output components, and adds a critical behavioral note all in one concise statement with no filler.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
With an output schema present and zero parameters, the description sufficiently covers the tool's purpose and behavior. The mention of 'sigma never toggles retrieval' addresses a potential concern, and the read-only diagnostic label sets accurate expectations. No significant gaps are apparent.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The tool has zero parameters, so the baseline score is 4. The description needs no parameter explanations; it appropriately focuses on output fields instead.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly identifies the tool as providing a snapshot of memory-graph topology, listing the specific fields (N, C, L, sigma, community_count, regime). It differentiates itself from sibling memory tools by being a read-only diagnostic, not a retrieval or mutation tool.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description explicitly labels it as a 'read-only diagnostic', making the intended use case (inspecting topology) clear. It does not mention explicit alternatives, but the context is sufficient given the tool's simplicity and zero parameters.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections. Dates show when Glama detected each change.
v3.1.0claim_checkmemory_capture9 fields changedInput schema / properties / agent_complete_idAdded value: +{
+ "description": "Optional id of a previously spawned background agent to mark complete on this call.",
+ "type": "string"
+}Input schema / properties / agent_expected_artifactAdded value: +{
+ "description": "Optional artifact the spawned background agent is expected to produce. Required alongside agent_id and agent_role to register a spawn; omitted otherwise.",
+ "type": "string"
+}Input schema / properties / agent_idAdded value: +{
+ "description": "Optional id of a background agent this capture is spawning or completing. Combine with agent_role and agent_expected_artifact to register a pending agent; combine with agent_complete_id on a later call to mark it done.",
+ "type": "string"
+}Input schema / properties / agent_modelAdded value: +{
+ "description": "Optional model label for the spawned background agent, recorded on the registry entry when agent_id/agent_role/agent_expected_artifact register a spawn.",
+ "type": "string"
+}Input schema / properties / agent_roleAdded value: +{
+ "description": "Optional role of the spawned background agent (for example 'research' or 'implement'). Required alongside agent_id and agent_expected_artifact to register a spawn; omitted otherwise.",
+ "type": "string"
+}Input schema / properties / epistemic_statusAdded value: +{
+ "default": "unknown",
+ "description": "Caller-declared epistemic status. Omit for 'unknown' (default, no behavior change). A value outside the enum is coerced to 'unknown' server-side, never rejected.",
+ "enum": [
+ "fact",
+ "estimate",
+ "hypothesis",
+ "opinion",
+ "unknown"
+ ],
+ "type": "string"
+}Input schema / properties / focusAdded value: +{
+ "description": "Optional current point of attention for the live session task. Folded verbatim onto this session's own working-tier entry after the capture completes, alongside next_action.",
+ "type": "string"
+}Input schema / properties / next_actionAdded value: +{
+ "description": "Optional immediate next step for the current live session task. Folded verbatim onto this session's own working-tier entry after the capture completes; surfaces at the next session start and on every subsequent turn until updated again.",
+ "type": "string"
+}Input schema / properties / salience_levelAdded value: +{
+ "default": "unflagged",
+ "description": "Caller-declared salience level for a decision, correction, or load-bearing preference marked in-turn. Additive rank-fusion boost only -- never a merge/drop lock. Omit for 'unflagged' (default, no behavior change). A value outside the enum is coerced to 'unflagged' server-side, never rejected.",
+ "enum": [
+ "unflagged",
+ "notable",
+ "critical"
+ ],
+ "type": "string"
+}memory_contradict1 field changedInput schema / properties / epistemic_statusAdded value: +{
+ "default": "unknown",
+ "description": "Caller-declared epistemic status of the corrected fact. Omit for 'unknown' (default, no behavior change). A value outside the enum is coerced to 'unknown' server-side, never rejected.",
+ "enum": [
+ "fact",
+ "estimate",
+ "hypothesis",
+ "opinion",
+ "unknown"
+ ],
+ "type": "string"
+}memory_reinforce1 field changedInput schema / properties / session_idAdded value: +{
+ "description": "Session identifier for correlating this reinforcement with the session's retrieval history. Optional; omit for old clients.",
+ "type": "string"
+}v3.0.3camouflaging_statusv3.0.0camouflaging_statuscuriosity_pendingepisodes_recentevents_querymemory_capturememory_consolidatememory_contradict1 field changedInput schema / properties / cue_embedding / descriptionPrevious value: -"Optional pre-computed embedding vector for the contradicting fact (its dimension must match the current store). When omitted, the daemon embeds new_fact server-side."New value: +"Optional pre-computed embedding vector for the contradicting fact (EMBED_DIM=384 floats; bge-small-en-v1.5). When omitted, the daemon embeds new_fact server-side."memory_recallmemory_recall_structuralmemory_reinforcememory_searchmemory_temporal_recallprofile_get_setschema_listtopologyv2.6.1camouflaging_statuscuriosity_pendingepisodes_recentevents_querymemory_capturememory_consolidatememory_recallmemory_recall_structuralmemory_reinforcememory_searchmemory_temporal_recallprofile_get_setschema_listtopologyv2.3.1memory_contradict1 field changedInput schema / properties / cue_embedding / descriptionPrevious value: -"Optional pre-computed embedding vector for the contradicting fact (EMBED_DIM=384 floats; bge-small-en-v1.5). When omitted, the daemon embeds new_fact server-side."New value: +"Optional pre-computed embedding vector for the contradicting fact (its dimension must match the current store). When omitted, the daemon embeds new_fact server-side."memory_recall2 fields changedInput schema / properties / cue / descriptionPrevious value: -"Natural-language query to match against stored memories. Embedded server-side via bge-small-en-v1.5 (384d) unless `cue_embedding` is supplied."New value: +"Natural-language query to match against stored memories. Embedded server-side by the configured provider unless `cue_embedding` is supplied."Input schema / properties / cue_embedding / descriptionPrevious value: -"Optional pre-computed embedding vector for the cue (EMBED_DIM=384 floats; bge-small-en-v1.5). When omitted, the daemon embeds the cue server-side. Used by memory_contradict and tests that need byte-stable embeddings."New value: +"Optional pre-computed embedding vector for the cue (its dimension must match the current store). When omitted, the daemon embeds the cue server-side. Used by memory_contradict and tests that need byte-stable embeddings."v2.0.0memory_searchmemory_temporal_recalltopology4 fields changedOutput schema / properties / C / typePrevious value: -"number"New value: +[
+ "number",
+ "null"
+]Output schema / properties / L / typePrevious value: -"number"New value: +[
+ "number",
+ "null"
+]Output schema / properties / rich_club_ratio / typePrevious value: -"number"New value: +[
+ "number",
+ "null"
+]Output schema / properties / sigma / typePrevious value: -"number"New value: +[
+ "number",
+ "null"
+]v1.0.3memory_capture1 field changedInput schema / properties / session_id / descriptionPrevious value: -"Current session id for provenance (MEM-05)."New value: +"Current session id for provenance."memory_recall1 field changedInput schema / properties / session_id / descriptionPrevious value: -"Current session id; gets written into every recalled record's provenance (MEM-05). Omit to use '-'."New value: +"Current session id; gets written into every recalled record's provenance. Omit to use '-'."episodes_recentmemory_recall_structural1 field changedInput schema / properties / structure_query / descriptionPrevious value: -"Optional role->filler map, e.g. {\"agent\": \"alice\"}. Each value is hashed to a filler hypervector. When omitted or empty, query HV is zero-filled and every row with structure_hv is scored (expensive at large N)."New value: +"Optional role->filler map, e.g. {\"agent\": \"agent_name\"}. Each value is hashed to a filler hypervector. When omitted or empty, query HV is zero-filled and every row with structure_hv is scored (expensive at large N)."Several tools are recall-like and could be confused: memory_search, memory_recall, claim_check, memory_recall_structural, memory_temporal_recall, and episodes_recent. The descriptions add strong usage cues and preferences, but the number of overlapping retrieval modes still creates meaningful misselection risk.
The memory_ prefix unifies core operations, but other tools use inconsistent styles: claim_check, schema_list, events_query, curiosity_pending, topology, episodes_recent. Modifier placement also varies between memory_recall_structural and memory_temporal_recall, so the set is readable but not patterned.
15 tools is at the upper edge of a reasonable scope, but each tool maps to a distinct memory function: capture, recall variants, consolidation, schemas, events, topology, and profile control. A few recall modes could arguably be parameterized into one tool, but the count is not excessive for the domain.
Core memory lifecycle is covered: capture, recall, contradiction, reinforcement, consolidation, diagnostics, and profile control. There is no direct get-by-id or update-record operation, and search hints may require follow-up verification, but the append-only design and diagnostic tools fill most agent workflows.
Cloud-hosted MCP server for durable AI memory
Persistent, portable memory for AI assistants — your private memory graph, from any MCP client.
An MCP memory server. One memory your agents share — across models, devices and apps.
Persistent personal memory for AI assistants — save, search, and recall across every MCP client.
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/CodeAbra/iai-personal-memory-engine'
If you have feedback or need assistance with the MCP directory API, please join our Discord server
