Skip to main content
Glama

Aquedra Run

Chat with an open model

chat

Answers a chat with open models on Aquedra Run's own GPUs: qwen3-8b (reasons before answering, and those tokens count as output and use up max_tokens, so set max_tokens to at least 1024); qwen2.5-3b (small text model for simple tasks); qwen2.5-coder-7b (code); qwen2.5-vl-7b (reads data: images). It runs only on idle GPUs: if the GPUs are busy the call fails with gpu_busy and retry_after_seconds, and nothing is charged. Limits are 8192 tokens of context, up to 2048 output tokens (a larger max_tokens is lowered to that, and the result then has max_tokens_requested) and 1 MiB of arguments, with no tools, JSON mode or embeddings; qwen2.5-vl-7b takes up to 2 inline data: images of up to 512 KB each, which together must fit in the 1 MiB. Costs per million tokens (input/output): qwen3-8b US$0.18/US$0.70, qwen2.5-3b US$0.05/US$0.10, qwen2.5-coder-7b US$0.025/US$0.05, qwen2.5-vl-7b US$0.18/US$0.69; minimum US$0.0005 per call, taken from prepaid credits and charged only when the call works; prompts and answers are not stored. Set dry_run to true to get the maximum a call can cost and max_cost_usd to refuse a call that could cost more.

Tool call template

Replace each placeholder with a value matching the input schema. Only required parameters are included; add optional parameters as needed.

{
  "id": 1,
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "arguments": {
      "messages": "<messages>"
    },
    "name": "chat"
  }
}

Input Schema

NameRequiredDescriptionDefault
nNoOnly 1 is supported.
seedNoInteger seed for repeatable sampling.
stopNoUp to 4 stop sequences of up to 64 characters each.
modelNoModel id, default qwen3-8b. See GET /v1/models for the ones on sale.
top_pNoNucleus sampling, 0 to 1.
dry_runNoWhen true, return the most this call can cost (cost_usd, the hold) and the max_tokens it assumes, without running it or charging anything.
messagesYesThe conversation, 1 to 64 messages.
max_tokensNoMost tokens to generate, default 1024. A larger value than the output cap (max_output_tokens in GET /v1/models) is lowered to it, and the response says so in max_tokens and max_tokens_requested. qwen3-8b's reasoning counts toward it: send at least 1024.
temperatureNoSampling temperature, 0 to 2.
max_cost_usdNoThe most this call may cost, in USD, as a number or decimal text with at most 6 decimals (e.g. 0.01 or "0.01"). A call that would cost more is refused with 402 max_cost_exceeded and nothing is charged.
response_formatNoOnly {"type":"text"} is supported.
max_completion_tokensNoSame as max_tokens; wins when both are sent.

Output Schema

NameRequiredDescriptionDefault
modelYesThe model that answered (or would answer, with dry_run).
priceNoDry run only: the model's price per million tokens.
usageNoTokens the call used.
objectNo
call_idNoServer-issued id of the call.
contentNoThe model's answer; absent with dry_run.
dry_runNoTrue when nothing was run or charged.
cost_usdYesWhat the call cost (or the most it can cost, with dry_run), US dollars as decimal text.
max_tokensNoThe max_tokens used: always with dry_run (the one the quote assumes); otherwise only when the max_tokens asked for was above the output cap and was lowered to it.
balance_usdNoCredit balance after the charge, US dollars as decimal text.
min_call_usdNoDry run only: the minimum charge per call.
finish_reasonNostop, or length when max_tokens ran out.
max_tokens_requestedNoOnly when max_tokens was lowered to the output cap: the max_tokens asked for.
estimated_prompt_tokensNoDry run only: the estimate of the prompt's tokens.

Schema Changelog

Changes observed during successful MCP inspections.

  1. First observed

TDQS

A4.6/5.0
Behavior5/5

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

Annotations only declare readOnlyHint=false and destructiveHint=false; the description supplies the behavior an agent actually needs: GPU availability gating, the gpu_busy failure with retry_after_seconds, no charge on failure, prepaid-credit billing with a US$0.0005 minimum, per-million-token prices per model, and that prompts/answers are not stored. It also documents the 402 max_cost_exceeded refusal path, which directly explains the non-read-only annotation.

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 a single dense paragraph front-loaded with purpose and model roster, and nearly every clause carries operational weight (limits, prices, failure modes). The trade-off is readability: pricing and token caps are buried in semicolon-chained clauses that would scan faster as a list, but there is little genuinely redundant text to cut.

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 12-parameter, billing-affecting inference call this covers everything an agent needs before invoking: model choice, context and output caps, request size, multimodal constraints, unsupported features, cost exposure and the dry_run/max_cost_usd guardrails. An output schema exists, so return-value shape is correctly left to it.

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 already 100%, so the baseline is 3, but the description adds non-obvious semantics the schema cannot: qwen3-8b reasoning tokens consume max_tokens so at least 1024 must be sent, oversized max_tokens is silently lowered, and a 1 MiB argument cap applies across all inputs. The image constraints for qwen2.5-vl-7b (2 inline data: images, 512 KB each, within the shared 1 MiB) restate schema text rather than extending it, which keeps this at 4 rather than 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?

The opening sentence names a specific verb and resource ('Answers a chat with open models') and immediately enumerates the four served models with their roles (qwen3-8b reasoning, qwen2.5-3b simple text, coder-7b code, vl-7b images). Nothing about this overlaps with siblings like buy_credits or the convert_* tools, so an agent can route here 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 concrete selection guidance among the four models and a clear cost-safety workflow ('Set dry_run to true to get the maximum a call can cost and max_cost_usd to refuse a call that could cost more'), plus the condition under which the call fails (busy GPUs). It stops short of explicit when-not-to-use routing beyond 'no tools, JSON mode or embeddings', but the context is unambiguous.

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

Try in Browser

Glama MCP Gateway

Add one secure layer between your agents and this server.

Resources