API Reference - Veri Documentation

Base URL

https://api.veri.studio/v1

Authentication

All API requests require a Bearer token. Create an API key from the dashboard or via the API keys endpoint.

curl https://api.veri.studio/v1/training_jobs \
  -H "Authorization: Bearer vk_your_api_key"

API keys use the vk_ prefix and are hashed server-side. Manage keys from the Veri dashboard.

Endpoints

These tables cover the core resources. Every endpoint, including evals, agents, webhooks, secrets and models, has its own page in this API reference with the full request and response schema.

Datasets

Method	Endpoint	Description
`POST`	`/v1/datasets`	Upload a JSONL dataset
`POST`	`/v1/datasets/connect`	Register an external dataset source
`GET`	`/v1/datasets`	List datasets
`GET`	`/v1/datasets/{id}`	Get dataset info

Training Jobs

Method	Endpoint	Description
`POST`	`/v1/training_jobs`	Create a training job
`GET`	`/v1/training_jobs`	List training jobs (filter by `status`, `method`)
`GET`	`/v1/training_jobs/{id}`	Get job status
`POST`	`/v1/training_jobs/{id}/cancel`	Cancel a job
`DELETE`	`/v1/training_jobs/{id}`	Hard-delete a finished job
`GET`	`/v1/training_jobs/{id}/logs`	Stream stdout tail
`GET`	`/v1/training_jobs/{id}/logs/full`	Full stdout from S3 (after job ends)
`GET`	`/v1/training_jobs/{id}/events`	Get job lifecycle events
`GET`	`/v1/training_jobs/{id}/metrics`	Sampled training metrics
`GET`	`/v1/training_jobs/{id}/model`	Get checkpoint download URL

See the Training jobs API for full schemas.

Deployments

Method	Endpoint	Description
`POST`	`/v1/deployments`	Create an inference deployment
`GET`	`/v1/deployments`	List deployments (filter by `status`)
`GET`	`/v1/deployments/{id}`	Get deployment details
`POST`	`/v1/deployments/{id}/stop`	Stop a deployment
`POST`	`/v1/deployments/{id}/chat/completions`	OpenAI-compatible inference
`GET`	`/v1/deployments/{id}/requests`	List per-request logs
`GET`	`/v1/deployments/{id}/metrics`	Get aggregate metrics

Volumes

See the Volumes guide for usage.

Method	Endpoint	Description
`POST`	`/v1/volumes`	Create a volume
`GET`	`/v1/volumes`	List volumes
`GET`	`/v1/volumes/{name}`	Get a volume
`DELETE`	`/v1/volumes/{name}`	Delete a volume
`GET`	`/v1/volumes/{name}/files`	List files
`DELETE`	`/v1/volumes/{name}/files/{path}`	Delete a file
`POST`	`/v1/volumes/{name}/upload-url`	Get a presigned upload URL

{name} accepts the volume’s name or its id (the one POST /v1/volumes returns).

Billing

Method	Endpoint	Description
`GET`	`/v1/billing/balance`	Get credit balance
`GET`	`/v1/billing/transactions`	List balance transactions

API Keys

Method	Endpoint	Description
`POST`	`/v1/api_keys`	Create a new API key (plaintext returned once)
`GET`	`/v1/api_keys`	List active API keys (prefix only)
`DELETE`	`/v1/api_keys/{id}`	Revoke an API key

Request Format

JSON endpoints (POST /v1/training_jobs): Use Content-Type: application/json with a JSON body. Reward functions for GRPO jobs ride this body as source text (reward_source), not as a file upload.
Upload endpoints (POST /v1/datasets, code artifacts): Use multipart/form-data.
Connector endpoints (POST /v1/datasets/connect): Send JSON bodies.

Response Format

Responses are standard JSON. Every response object includes an object field that identifies the resource type (e.g., "training_job", "deployment", "dataset", "api_key"). Single-resource endpoints return one object. List endpoints return cursor-paginated responses:

{
  "object": "list",
  "data": [...],
  "has_more": false
}

Success (2xx)

{
  "id": "3f9c2a7b1d40",
  "object": "training_job",
  "status": "queued",
  "base_model": "Qwen/Qwen3-4B",
  "method": "grpo",
  "output_name": "my-model",
  "gpu_requested": {
    "type": "L4-24GB",
    "count": 1
  },
  "gpu": null,
  "created_at": "2026-04-14T12:00:00Z"
}

Error (4xx/5xx)

{
  "error": {
    "message": "Invalid, revoked or expired API key"
  }
}

Errors a client can branch on also carry a machine-readable code, and retryable ones (429 rate limit, 503) add retry_after_seconds and a Retry-After header:

{
  "error": {
    "code": "RATE_LIMITED",
    "message": "rate limit exceeded (120 requests per 60s for this key); retry after the current window resets",
    "retry_after_seconds": 59
  }
}

Some 409 refusals add structured fields next to code (for example, the promotion gate lists the failing evaluators). A body that is not valid JSON, has a missing or mistyped field, or is sent without Content-Type: application/json is rejected before it reaches the endpoint, with a plain-text message instead of JSON (400, 422 or 415).

Common Status Codes

Status	Meaning
400	Invalid request: malformed JSON, a failed validation, or an unsupported state transition
401	Missing, invalid, revoked or expired API key, or a model access key used outside its deployment
402	Insufficient credit balance, or no billing account linked to the workspace yet
404	Resource not found or not in your workspace
409	Conflict: the name is taken, or the resource’s current state does not allow the action
415	JSON endpoint called without `Content-Type: application/json`
422	Request body has a missing or mistyped field
429	Rate limit (`RATE_LIMITED`: 120 requests per minute per key, 3,600 per minute for `chat/completions`; honor `Retry-After`), or a workspace concurrency cap: `CONCURRENT_JOB_LIMIT`, `ACTIVE_DEPLOYMENT_LIMIT`, `CONCURRENT_SESSION_LIMIT`, `CONCURRENT_EXPERIMENT_LIMIT` or `TRIGGER_LIMIT` (no `Retry-After`: finish or stop something first, or buy the Pro pack, see Plans)
500	Unexpected server error
501	The feature is not available on this server yet
503	Temporarily unavailable, for example a deployment waking from scale-to-zero (`deployment_waking`); retry after `Retry-After` seconds

Pagination

Most list endpoints support cursor-based pagination with limit and after query parameters:

curl "https://api.veri.studio/v1/datasets?limit=10&after=3f9c2a7b1d40" \
  -H "Authorization: Bearer vk_your_api_key"

Parameter	Type	Default	Description
`limit`	integer	20 (deployments: 50)	Maximum number of items to return
`after`	string	—	ID of the last item from the previous page (cursor)

The response includes a has_more field indicating whether additional pages exist:

{
  "object": "list",
  "data": [...],
  "has_more": true
}

To paginate, pass the id of the last item in data as the after parameter in the next request. A few short lists are returned whole, without a cursor (for example, webhook endpoints, up to 100). Each endpoint’s page lists the parameters it accepts.

​Base URL

​Authentication

​Endpoints

​Datasets

​Training Jobs

​Deployments

​Volumes

​Billing

​API Keys

​Request Format

​Response Format

​Common Status Codes

​Pagination