Concepts
A tool is a workflow definition published as a callable capability for external AI agents. Each published tool exposes the workflow's name, description, and input schema in a format compatible with the Anthropic and OpenAI tool use APIs.
The model
A tool has three fields:
- Name — the workflow handle. Stable across publish/unpublish cycles.
- Description — a human-readable summary of what the tool does.
- Input schema — a JSON Schema Draft 7 object describing the inputs the workflow accepts.
Publish a workflow as a tool from the workflow editor. Invoking a tool starts a workflow run.
The invocation waits synchronously up to timeout_seconds (default 30 s, max 120 s). When the
run completes within that window, the response carries the run output. When it does not, the
response includes the run ID and status: pending; poll for the result using the run ID.
Invoking a tool from an agent
curl -X POST "$MOBIUS_BASE_URL/v1/projects/acme/tools/summarize-document/runs" \
-H "Authorization: Bearer $MOBIUS_API_KEY" \
-H "Content-Type: application/json" \
-d '{"input": {"doc_id": "doc_abc123"}, "timeout_seconds": 60}'Mobius starts the summarize-document workflow with inputs.doc_id set to doc_abc123 and
waits up to 60 seconds. A 200 response includes the completed run output. A 202 response
means the run did not finish in time; poll
GET /v1/projects/acme/tools/summarize-document/runs/{run_id} until status is completed or
failed.
Lifecycle
A tool run moves through these states:
| State | Entered when | Left when |
|---|---|---|
pending | Run is created but not yet executing | First step begins |
running | First step begins | All steps finish or a step fails |
completed | All steps finish successfully | — |
failed | A step fails without recovery | — |
suspended | Run pauses waiting for a signal or interaction | Run resumes |
Where you see it
- Dashboard — Projects → <project> → Tools lists every published tool with its description and input fields.
- API — the Tools tag group covers
listTools,runTool, andgetToolRun.