Which agents work with it?

Any agent that reads skills from a directory. Claude Code, Codex and Gemini CLI read it natively; Cursor and VS Code connect over MCP.

The voice runs on your own machine, so there is nothing to meter. The first run downloads it, about 340 MB.

Yes. The skill is MIT-licensed and the authoring format is public.

Errors · Bisque docs

Every error is JSON with a stable code, a message for people, and a requestId to quote.

The error envelope

{
  "requestId": "…",
  "error": {
    "code": "INVALID_PRESENTATION_ID",
    "message": "presentationId query parameter is required.",
    "hint": "…"
  }
}

error.code is upper snake case and stable; match on it.
error.message is for people and can change.
error.hint, when there is one, says where to look next.
requestId is what to quote in a bug report.

Only error, with its code and message, is on every error; requestId and hint are present when the endpoint has them. The OAuth endpoints answer with RFC 6749 bodies instead, {"error": "…", "error_description": "…"}.

Common codes

Status	Code	Meaning
`401`	`UNAUTHORIZED`, `INVALID_ACCESS_TOKEN`	The credential is missing, expired, or revoked. See Authentication.
`404`	`NOT_FOUND`	No endpoint at that `/api` path. The body's `hint` names the OpenAPI document and the MCP server, and `docs` names the API overview.
`404`	`PRESENTATION_NOT_FOUND`	No presentation at that address, or one you cannot read.
`405`	`METHOD_NOT_ALLOWED`	The endpoint exists, but not for that HTTP method.
`422`	`IDEMPOTENCY_KEY_REUSED`	The same `Idempotency-Key` with a different body.
`429`	`CONTEXT_RATE_LIMITED`	The per-IP window on `/api/presentations/context`. Wait `Retry-After` seconds.

The live creation endpoints under /v1/presentations add their own. In every case nothing was written.

Status	Code	Meaning
`400`	`LIVE_APPEND_ONLY`	An append removed, reordered, or inserted a slide before one already sent. The message names the first slide that moved.
`404`	`NO_OPEN_CREATION`	Nothing is open under that id. A finalized presentation answers this.
`409`	`LIVE_TARGET_CHANGED`	An append sent a `slug`, `handle`, `org`, or `visibility` different from the one the creation started with. Leave them out and the originals are used.
`409`	`LIVE_REVISION_CONFLICT`	The `baseRevision` you sent is behind the current `live.revision`. Read the presentation, then append after what is there.
`409`	`LIVE_APPEND_IN_PROGRESS`	Another append or narration upload was still being written after 20 seconds. Try again.
`409`	`CREATION_SUPERSEDED`	Another publish of the same presentation replaced the open creation.

Rate limits

GET and POST /api/presentations/context are limited per IP: 60 requests per 10 minutes without a credential, 300 with one. Every response from the endpoint carries RateLimit-Limit, RateLimit-Remaining, RateLimit-Reset (seconds) and RateLimit-Policy (60;w=600 without a credential, 300;w=600 with one). A 429 adds Retry-After. In the batch form each item costs one slot, and items past the window come back as CONTEXT_RATE_LIMITED rows rather than failing the request.

The documentation MCP server, POST /docs/mcp, allows 120 requests per 10 minutes per IP.

Publishing endpoints have no request window.

Back to the API overview. For the tool-level view of the same API, see the MCP servers.