Access
Falcon is in private preview. What a preview key covers: the public routes for every series, key and session-token prefixes, quota buckets, and weights.
Every model can be tried without a credential at chat.falconlab.app: pick a model, write a message, record a clip or describe a scene, and read the answer. The chat site calls the same routes on a shared public account with per-visitor limits, so it is a place to look, not a place to build; the API below is.
Falcon is in private preview. Self-serve API keys and accounts are paused. Preview access is granted by an administrator, who issues portal credentials tied to an organisation; with them you sign in to the portal, create API keys, read usage against quota and try every route in the sandbox. The Portal page walks through it.
The Falcon API is the public HTTP surface for every series: nine routes under /v1/ — two for the LIM models, two for the Waymark models, two for the Response models and three for the Express models — one response envelope, bearer authentication and monthly quota buckets. This page describes what a preview credential covers. API conventions describes the envelope, error codes and retry behaviour shared by every route.
What a preview key covers #
A preview key is a string of the form fln_<id>_<secret>. Every request carries it as a bearer token:
POST /v1/intention
Authorization: Bearer fln_<id>_<secret>
Content-Type: application/jsonA key covers the public routes below. All eleven serve today:
| Route | Model | Bucket | Body limit |
|---|---|---|---|
POST /v1/intention | ELIM (default), LIM with model: "lim" or LIM Nano with model: "lim-nano" | intention | 64,000 bytes |
POST /v1/intention3d | LIM3D-XL (default), or LIM3D with model: "lim3d" | intention | 256,000 bytes |
POST /v1/spatial | Waymark, from a sketch or from an image or video sketched server-side | spatial | 24 MB (image data up to 6 MB, video up to 15 MB) |
POST /v1/move | Waymark Extra, from a sketch and a goal | spatial | 64,000 bytes; sketch ≤ 6,000 characters, goal ≤ 300 |
POST /v1/airspace/read | Waymark Flight, an airspace reading beside the exact check — decision support only | spatial | 64,000 bytes; ≤ 24 volumes, 1–12 track points |
POST /v1/airspace/plan | Waymark Flight, a route validated by the exact checker — decision support only | spatial | 64,000 bytes; ≤ 24 volumes |
POST /v1/chat | Cognitio | complete | 64,000 bytes; text ≤ 2,000 characters |
POST /v1/sound | Express Cue — released | audio | text ≤ 400 characters; byte limit fixed with the route |
POST /v1/speak | Express Voice — released; seven voices, lj the default | audio | text ≤ 600 characters; byte limit fixed with the route |
POST /v1/transcribe | Express2Text — released | audio | one WAV, FLAC or OGG clip of ≤ 60 s and ≤ 12 MB decoded, as base64 |
GET /v1/voices | The voice catalogue: lj, female-1 to female-3, male-1 to male-3 | none | — |
A route whose model has no hosted endpoint yet answers 503 with a model-specific unavailable code until the deployment exists; Status & versioning explains what that means. /v1/spatial and /v1/move share a GPU service that scales to zero, so the first call to either after a quiet period can take one to two minutes; /v1/chat scales to zero as well, and so does the Waymark Flight CPU service behind /v1/airspace/read and /v1/airspace/plan, whose first call waits a few seconds (503 flight_warming, Retry-After 20 s). The /v1/speak voice service and the Falcon API itself are kept warm with one minimum instance each, so the voice routes and the in-process models carry no start-up wait. The Express Cue service scales to zero after a quiet period and answers 503 while it loads.
A preview account session is identified by a token of the form fls_<id>_<secret>, which the same routes accept in place of a key. A session token expires after 30 days and can be revoked; a key does not expire but can be revoked, and an account holds at most eight keys. Keys are created and revoked in the portal; each is shown once, when it is created.
Preview quota buckets #
Each successful call counts once against one of four preview quota buckets, per UTC calendar month. Two allowance levels exist: an account that has only ever used session tokens counts against the smaller allowance; once a preview key has been issued to the account, its keys and its sessions both count against the larger one.
| Bucket | Routes | Session-only account | Account with a preview key |
|---|---|---|---|
| intention | /v1/intention, /v1/intention3d | 300 | 10,000 |
| spatial | /v1/spatial, /v1/move, /v1/airspace/read, /v1/airspace/plan | 60 | 2,000 |
| complete | /v1/chat | 100 | 5,000 |
| audio | /v1/sound, /v1/speak, /v1/transcribe | not yet fixed | not yet fixed |
When a bucket is exhausted, the route answers 429 and names the bucket:
{"ok": false, "error": "quota_exceeded", "kind": "spatial"}Counters reset at the start of the next UTC calendar month. A call that fails validation or authentication is not counted. The audio allowances have not been published yet; /v1/voices is not metered.
Need dedicated capacity? #
The shared fleet is what a preview key covers by default. An organisation that needs private, always-warm capacity can rent dedicated hosting by the month, and Web Hosting adds a static site at its own falconlab.app address; both are requested from the portal dashboard.
Weights #
Weights are not distributed during the private preview. Each model page records the format, size and identifier of its weights so that a deployment can be verified against the documentation, but the files themselves are not published. Fixed weights per version; the model does not learn from requests.
Contact #
Access to the preview is arranged by Ducky Software. An administrator creates the organisation and its users in the portal and hands each user a username and a one-time password; the user chooses a password of their own at first sign-in.
A contact route for preview access has not been published yet.