Ask Torque
Tool reference for the ask_torque escape-hatch tool
ask_torque proxies a natural-language question to the Torque AI assistant and returns its markdown reply inline. It is the catch-all tool for questions that do not map to a structured MCP tool.
For narrative, when-to-use guidance, and example prompts, see the Ask Torque guide.
ask_torque
Requires: authentication, active project.
| Parameter | Type | Required | Description |
|---|---|---|---|
question | string | Yes | The user's question, verbatim. Must be non-empty. Clarify ambiguous intent with the user first — vague prompts produce vague answers. |
additionalContext | string | No | Optional session context to forward (a file the user shared, a stack trace, output from an earlier tool). Passed to the assistant as userContext.additional_context. |
Return shape
A single text content block containing markdown — the assistant's reply, optionally followed by a **Data used:** section summarizing any queries the assistant ran. If the assistant reports an error, isError: true is set; the text still contains the error message.
Limits
- No chart rendering. The assistant's chart tool is suppressed in MCP. Data appears inline as markdown pipe-tables or Unicode sparkline rows.
- No suggestion chips. The web-UI follow-up prompt buttons are suppressed.
- Conversation cap. Per-session history is capped at ~20 messages; oldest user/assistant pairs drop first.
- Auto-reset on project switch. Switching the active project clears the conversation so context does not leak across projects. Call
reset_contextto force a fresh thread without switching.
When to reach for a different tool
ask_torque is the escape hatch. When the task maps to a structured tool, prefer that — it's faster, cheaper, and more reliable. See the routing table in the Ask Torque guide.