API reference

Every call goes to your issued base URL — https://api.verdictlabs.eu.org/…, path included. All endpoints speak JSON.

Overview

Endpoint	Key	Purpose
`POST /v1/decisions`	optional	Ask questions about one input.
`GET /v1/models`	no	The served models.
`GET /api/info`	no	Models, limits and service metadata as JSON.
`GET /api/status`	no	Service state and latency.
`GET /api/stats`	no	Aggregate usage counters.
`POST /api/keys/challenge`	no	Start key creation (returns a proof-of-work challenge).
`POST /api/keys`	no	Finish key creation.
`GET /api/keys/self`	yes	Your key's status and usage.
`DELETE /api/keys/self`	yes	Revoke your key.

Authentication

No key is required. Your base URL carries the access path; treat that URL as the credential and do not publish it. Calls made without the path return the same 404 as any unknown route.

If you do send a key it must be valid; an unknown or revoked key is rejected with 401.

Keys are optional and free: created on this site in one click, without an account, or from a terminal with the two proof-of-work calls below. Only a hash of each key is stored; a lost key cannot be recovered — create another one.

POST /api/keys/challenge      -> {"challenge":"...","difficulty":20,"expires_at":"..."}
POST /api/keys                -> {"challenge":"...","nonce":"...","accept_terms":true}
                              <- {"api_key":"sk-vrd-...","key_id":"...","created_at":"..."}

Solve the challenge by finding a decimal nonce such that SHA-256 of challenge + ":" + nonce starts with difficulty zero bits.

Request format

Field	Type	Meaning
`model`	string	One of the model IDs below.
`state`	string	The input the questions are about.
`questions`	object	Question IDs, each with its question. Answers return under the same IDs.
`images`	string[]	Optional base64 data URLs (PNG, JPEG, WebP).
`videos`	string[]	Optional base64 data URLs (MP4, WebM, QuickTime, Matroska).
`states`	object	Optional. Several inputs in one call, each with its own state and questions.

Ask everything you need about one input in a single request: the input is read once and every question is answered together.

Question types

Type	Asks	Fields	Answer
`choice`	Which option fits?	`criteria`: 2 or more options	`choice`, `probabilities`, `confidence`
`noul`	Yes or no?	optional `criteria`: {false, true} descriptions	`noul`: probability of yes
`score`	How much, on an ordered scale?	`criteria`: 2 to 10 levels, lowest first	`score`: expected level, `probabilities`
`box`	Where is it in the image?	none; the state carries one image	`box`: top, left, bottom, right, `confidence`

Every question carries type and instructions. Box questions are answered by the flagship model only.

Responses

{
  "answers": {
    "route": {"type":"choice","choice":"engineering",
              "probabilities":{"billing":0.0004,"engineering":0.9996,"sales":0.0000},
              "confidence":0.9993},
    "urgent": {"type":"noul","noul":0.87}
  },
  "model": "vrd/decision-1",
  "usage": {"input_tokens": 141, "output_tokens": 0}
}

Decision models produce no text: output_tokens is always 0 and usage.input_tokens counts the input read by every question.

Models

ID	Name	Profile	Inputs
`vrd/decision-1`	decision-1	flagship	text, image, video

Limits

No quotas on keys. Keys have no per-minute budget, no daily quota and no concurrency cap. Calls are served without a key; the only usage ceiling is a shared budget of 6.9M input tokens per minute — and more than a normal client can absorb, so a 429 is the ceiling anyway. Exceeding it returns retry-after in seconds; honour it and do not hot-loop. The request timeout is 180 seconds.

Request-size limits are technical, not usage limits:

Body	64 MB
Questions per request	64
Inputs (`states`)	8
Images per request	4
Videos per request	1 (60 s)
Characters per `state`	32,000

Errors

{"error":{"type":"invalid_request","code":"model_unknown","message":"..."}}

HTTP	Type	When
400	`invalid_request`	Malformed JSON, missing model or questions, unknown model, invalid question.
401	`unauthorized`	An unknown or revoked key was sent.
413	`request_too_large`	Body over the size limit.
429	`rate_limited`	The service is over its token budget; honour `retry-after`.
503	`service_busy`	The service is saturated; retry shortly.
504	`timeout`	The model did not finish in time.

Support

support@verdictlabs.eu.org