Docs

POST /v1/chat/completions — OpenAI-shaped chat with streaming, tools, and provider routing.

Chat completions

POST /v1/chat/completions

OpenAI-compatible. Usage-billed to the calling key and org credits. This is the path Playground uses.

Base: https://opendoor-gcp.web.app. Auth: Authorization: Bearer $OPENDOOR_API_KEY.

Headers

Header	Required	Description
`Authorization`	Yes	`Bearer $OPENDOOR_API_KEY`
`Content-Type`	Yes	`application/json`
`HTTP-Referer` / `X-Title`	No	Optional app attribution (logged on the request)
`x-opendoor-service-tier`	No	`standard` or `priority` if not set on the body

Body

Field	Type	Required	Description
`model`	string	Yes	Catalog id. Start with `gemma-4-26b-a4b-it` or an id from `GET /v1/models`
`messages`	array	Yes	`{ role, content }` — `content` is a string or multimodal parts
`stream`	boolean	No	SSE when `true`
`max_tokens`	integer	No	Completion cap (estimate uses 1024 if omitted)
`temperature`	number	No	0–2
`top_p`	number	No	Nucleus sampling
`frequency_penalty`	number	No	−2 to 2
`presence_penalty`	number	No	−2 to 2
`tools`	array	No	OpenAI-shaped tools
`tool_choice`	string/object	No	`auto`, `none`, `required`, or a named function
`response_format`	object	No	`json_object` or `json_schema`
`user`	string	No	End-user id (also used in prompt-cache fingerprint)
`service_tier`	string	No	`standard` (default) or `priority` (Pro / Team / Enterprise)
`prompt_cache_key`	string	No	Sticky cache key; the gateway sets one if omitted
`provider`	object	No	OpenRouter-shaped routing (below)
`transforms`	array	No	Message transforms (context window)

Pick a live catalog id. Do not treat coming_soon as the way to chat — hosted Vertex MaaS ids such as gemma-4-26b-a4b-it, qwen3-next-80b-instruct, and deepseek-v3.2 are callable when the wholesale path is configured.

Example

bash
export OPENDOOR_API_KEY=opd_…
export OPENDOOR_BASE_URL=https://opendoor-gcp.web.app

curl "$OPENDOOR_BASE_URL/v1/chat/completions" \
  -H "Authorization: Bearer $OPENDOOR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemma-4-26b-a4b-it",
    "messages": [
      {"role": "system", "content": "You are a helpful assistant."},
      {"role": "user", "content": "What is OpenDoor?"}
    ],
    "max_tokens": 200
  }'

Successful responses set id (and x-generation-id) to the logged generation UUID. Fetch it with GET /v1/generation/:id (also mounted at /v1/generations/:id).

json
{
  "id": "generation-uuid",
  "object": "chat.completion",
  "created": 1777777777,
  "model": "gemma-4-26b-a4b-it",
  "choices": [
    {
      "index": 0,
      "message": { "role": "assistant", "content": "…" },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 20,
    "completion_tokens": 40,
    "total_tokens": 60
  }
}

Streaming

bash
curl "$OPENDOOR_BASE_URL/v1/chat/completions" \
  -H "Authorization: Bearer $OPENDOOR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemma-4-26b-a4b-it",
    "messages": [{"role":"user","content":"Hello"}],
    "stream": true
  }'

data: {"id":"…","object":"chat.completion.chunk","choices":[{"delta":{"content":"Hello"}}]}

data: [DONE]

Provider routing

Field	Type	Default	Description
`order`	string[]	—	Try these slugs first (`vertex`, `together`, `openai`, …)
`allow_fallbacks`	boolean	`true`	If false, only the first remaining provider is tried
`sort`	`price` \| `latency` \| `throughput`	—	Re-rank the candidate chain
`only`	string[]	—	Restrict to these slugs
`ignore`	string[]	—	Drop these slugs

Org BYOK slugs (and alwaysUse keys) are preferred before platform env keys. Data-residency filters still apply.

bash
curl "$OPENDOOR_BASE_URL/v1/chat/completions" \
  -H "Authorization: Bearer $OPENDOOR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemma-4-26b-a4b-it",
    "messages": [{"role":"user","content":"Hello"}],
    "provider": { "order": ["vertex"], "allow_fallbacks": true }
  }'

JavaScript / Python

ts
import { OpenDoor } from "@opendoor/sdk";

const client = new OpenDoor({
  apiKey: process.env.OPENDOOR_API_KEY,
  baseURL: "https://opendoor-gcp.web.app",
});

const chat = await client.chat.completions.create({
  model: "gemma-4-26b-a4b-it",
  messages: [{ role: "user", content: "Hello" }],
});

python
from opendoor import OpenDoor

client = OpenDoor(base_url="https://opendoor-gcp.web.app")
out = client.chat.completions.create(
    model="gemma-4-26b-a4b-it",
    messages=[{"role": "user", "content": "Hello"}],
)

Errors

See Errors for 401 / 402 / 403 / 404 / 429 / 502 / 503 bodies, including insufficient balance and “all providers failed”.

API keys

Models