List Agent Tools

List all tools available to the Talonic embedded AI agent with impact levels from read to irreversible. Build permission-aware integrations and confirmation gates.

The agent tools endpoint returns the agent's live tool registry: it is read directly from the same registry the agent loop executes from, so the list can never drift from what actually runs. Each entry carries a stable name, a human-readable description, an impact level that classifies how consequential the operation is, a capability tag naming the permission the tool requires, a can_invoke flag stating whether the calling key can run the tool directly, and the tool's input_schema, a JSON Schema describing its arguments. Use it for tool discovery: dynamically generating tool descriptions for external AI agents, auditing available capabilities, or building permission-aware integrations.

Call this endpoint at startup to populate your integration's tool registry, or periodically to detect newly added capabilities: because the response reflects the live loop, a newly shipped tool appears here the moment it is deployed, with no docs lag. The input_schema on each entry is what makes direct invocation programmable: generate argument forms or function-call definitions from it, then run the tool through [POST /v1/agent/tools/:name/invoke](agent-invoke-tool).

The totalCount field gives the total number of registered tools, and invocable_count how many of them this key can invoke. The registry lists every tool the in-product agent operates, 61 at present, while an API key runs as the least-privilege viewer role and can invoke only the read subset, currently 32; that asymmetry is exactly why can_invoke exists, so the listing is a menu of what you can order and a transparent view of the rest. Each tool's impact field follows a four-level severity scale: read, draft_mutation, live_mutation, and irreversible. Use these levels to build confirmation gates: for example, auto-approve read tools but require user confirmation for live_mutation and above. The capability field is the finer-grained permission the loop checks against the caller's role before executing (for example data.read, source.ingest, review.resolve, spec.author).

Pair this with the Workspace Context endpoint to give your external AI agent both situational awareness (context) and available actions (tools). The tool names returned here are stable identifiers that can be referenced in custom orchestration logic or permission policies.

Response

{
  "tools": [
    {
      "name": "query_data",
      "impact": "read",
      "capability": "data.read",
      "description": "Runs a read-only SQL query over the structured cell plane.",
      "can_invoke": true,
      "input_schema": { "type": "object", "properties": { "sql": { "type": "string" } } }
    },
    {
      "name": "promote_field",
      "impact": "draft_mutation",
      "capability": "data.promote",
      "description": "Makes an already-captured field queryable. No LLM call, reversible.",
      "can_invoke": false,
      "input_schema": { "type": "object", "properties": { "field_id": { "type": "string" } } }
    }
  ],
  "totalCount": 61,
  "invocable_count": 32
}

Errors