Inference API

Base URL: https://api.base.devmaagency.com/v1. All requests need an API key in the Authorization header.

Authentication

Send Authorization: Bearer bvx_…. Keys are created in the console and shown once. A key can be limited to one client and to specific deployments; calls outside those limits return deployment_not_allowed_for_key.

POST /v1/chat/completions

Generates a response from the adapter behind a deployment alias.

Field	Type	Notes
`model`	string	Required. Your deployment alias.
`messages`	array	Required. `system`, `user` and `assistant` messages with text content.
`max_tokens` / `max_completion_tokens`	integer	Optional. Clamped to the deployment’s maximum output tokens; defaults to that maximum.
`temperature`	number	Optional, 0–2.
`top_p`	number	Optional, 0–1.
`stop`	string or array	Optional, up to 4 sequences.
`stream`	boolean	Optional. Server-sent events; see below.
`stream_options.include_usage`	boolean	Optional. Adds a final chunk with token usage.
`seed`	integer	Optional. Best-effort determinism.

Any other field — including tools, functions, n greater than 1, logprobs, response_format and image content — returns 400 unsupported_parameter rather than being silently ignored.

Response

json

{
  "id": "chatcmpl-01a1843c...",
  "object": "chat.completion",
  "created": 1793220000,
  "model": "contract-review",
  "choices": [{
    "index": 0,
    "message": { "role": "assistant", "content": "Clause 4.2 limits liability to fees paid in the prior 12 months." },
    "finish_reason": "stop"
  }],
  "usage": { "prompt_tokens": 41, "completion_tokens": 17, "total_tokens": 58 }
}

Streaming

With "stream": true the response is text/event-stream, one JSON chunk per data: line, ending with data: [DONE].

text/event-stream

data: {"id":"chatcmpl-…","object":"chat.completion.chunk","choices":[{"index":0,"delta":{"content":"Clause"}}]}

data: {"id":"chatcmpl-…","object":"chat.completion.chunk","choices":[{"index":0,"delta":{"content":" 4.2"}}]}

data: [DONE]

If the client disconnects mid-stream, generation stops. Tokens the runtime already processed are recorded and billed; see Usage & billing.

GET /v1/models

Lists the deployment aliases your key can call, in the OpenAI list format ({"object": "list", "data": [{"id": "contract-review", "object": "model", …}]}). Paused deployments are not listed.

Response headers

Header	Meaning
`x-request-id`	Unique id for the request. Include it when contacting support.
`x-basevex-synthetic`	`true` when the response is synthetic (sandbox). Synthetic responses also have `"system_fingerprint": "basevex-synthetic"` and text starting with `[SYNTHETIC]`.

Errors

Errors use the OpenAI error shape, so SDKs raise them normally. See Error codes.