# Five Nine Agent

> Live catalogs from the open web, under signed receipts. Ask a natural-language
> question; get attributed rows, per-source signed receipts, and a durable,
> replayable session you can follow up on. Honest by construction: unchecked
> sources are stated, abstains are $0 outcomes, delivered work is itemised per source.
> Receipts itemise DELIVERED passes only: orchestration spend and passes that delivered
> nothing are billed without a per-source receipt, so sum(receipts) <= spend_usd and the
> ledger (GET /v1/usage), not the receipts sum, is the complete account of a bill.

API base: https://api.fiveninelabs.com/v1 (Bearer fn_live_/fn_test_ keys).
A real run takes tens of minutes (measured 11-68 min per session; first rows ~4-17 min).
NOT YET AVAILABLE (planned — marked x-fn-status: planned in the reference; each
answers 501 not_implemented with a `surface` field, never 404): /discover,
/extractions, /monitors, refresh, redaction (DELETE), /receipts/{id}, webhook
test/redeliver, and the MCP endpoint (/mcp).
A refusal to spend is a 201 -> $0 `abstained` run (pre-spend signal:
estimate.would_run:false) — never a 422.
Every page below also serves a .md variant at the same URL plus ".md".

## Docs

- [Quickstart](https://docs.fiveninelabs.com/quickstart.md): free estimate, instant $0 test-key replay, then the real run — with the wait announced
- [Ways to watch a run](https://docs.fiveninelabs.com/guides/streaming.md): SSE with resume, the rebuild-then-attach reload recipe, long-poll (timeout ≤ 240 + Retry-After loop), stop semantics (in-flight passes drain), webhooks
- [Browser clients](https://docs.fiveninelabs.com/guides/browser.md): CORS contract, why EventSource can't connect, the scoped-token pattern, never ship a live key to a page
- [Budgets & proposals](https://docs.fiveninelabs.com/guides/proposals.md): max_spend_usd as reservation and hard ceiling; the 409 priced-proposal confirm flow
- [Event vocabulary](https://docs.fiveninelabs.com/guides/events.md): the stable ProgressEvent subset regenerated from the live wire — unknown phases are liveness; the three row dialects mapped
- [Webhooks](https://docs.fiveninelabs.com/guides/webhooks.md): Standard Webhooks signing, retry schedule, full event catalog (test deliveries + redeliver are planned)
- [Test mode](https://docs.fiveninelabs.com/guides/test-mode.md): fn_test_ keys (how to get one), recorded-session replays, and the FN_TEST_* sentinels
- [Errors](https://docs.fiveninelabs.com/errors.md): the one envelope and every code — each code has its own page at /errors/<code>.md

## Planned surfaces (contracts published, endpoints not yet served)

- [Extractions](https://docs.fiveninelabs.com/guides/extractions.md): URL-first — discover, extract, batch; claims stated as complete or partial
- [Refresh & monitors](https://docs.fiveninelabs.com/guides/refresh-and-monitors.md): fresh data at marginal cost, $0 within max_age_s, standing monitors with webhook diffs (re-reading finished runs is live and free)
- [MCP setup](https://docs.fiveninelabs.com/guides/mcp.md): add the agent to your agent — tools, auth, config snippets

## Optional

- [Overview](https://docs.fiveninelabs.com/index.md)
- [Changelog](https://docs.fiveninelabs.com/changelog.md)
- [Dashboard](https://platform.fiveninelabs.com): keys, usage, session forensics, playground