Quickstart

POST /v1/chat/completions takes the same shape as OpenAI’s Chat Completions API — model and messages are the only required fields. Use whichever model id your Connector allows — check with whoever set up your Tenant if you’re not sure which ones those are.

curl -sS "<staging Gateway base URL — set NEXT_PUBLIC_GATEWAY_BASE_URL>/v1/chat/completions" \
  -H "Authorization: Bearer fwk_..." \
  -H "Content-Type: application/json" \
  -d '{
  "model": "<model id from your Connector'\''s allow-list>",
  "messages": [
    {
      "role": "user",
      "content": "Summarize this week'\''s open invoices."
    }
  ]
}'

A successful call returns:

{
  "id": "chatcmpl-01jazyb3x8k9q2m4n5p6r7s8t9",
  "object": "chat.completion",
  "created": 1758500000,
  "model": "<model id from your Connector's allow-list>",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "..."
      },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 24,
    "completion_tokens": 96,
    "total_tokens": 120
  }
}

A few request fields OpenAI’s API accepts aren’t supported yet and are rejected by name with a 400 rather than silently ignored — tools, tool_choice, response_format, n, stop, top_p, seed, logprobs, presence_penalty and frequency_penalty. Message content also has to be plain text — an image or file content part is rejected the same way, rather than silently dropped from the request that gets audited.

Add "stream": true to the same request and, once your Agent’s Connector supports it, the Gateway responds with Server-Sent Events instead of one JSON body — the same chat.completion.chunk shape the OpenAI SDK already knows how to consume, one data: {...} frame per chunk, closed by a literal data: [DONE]:

curl -sS -N "<staging Gateway base URL — set NEXT_PUBLIC_GATEWAY_BASE_URL>/v1/chat/completions" \
  -H "Authorization: Bearer fwk_..." \
  -H "Content-Type: application/json" \
  -d '{
  "model": "<model id from your Connector'\''s allow-list>",
  "messages": [
    {
      "role": "user",
      "content": "Summarize this week'\''s open invoices."
    }
  ],
  "stream": true
}'

data: {"id":"chatcmpl-01jazyb3x8k9q2m4n5p6r7s8t9","object":"chat.completion.chunk","created":1758500000,"model":"<model id from your Connector's allow-list>","choices":[{"index":0,"delta":{"content":"Sum"},"finish_reason":null}]}

data: ...

data: [DONE]

The Gateway strips two categories of value out of every call before it’s written to storage — automatically, on every Tenant, with nothing for you to configure today:

• Sensitive-looking field names. Any JSON key containing (case-insensitively, at any depth, in either the request or the response) one of: password, passwd, secret, api_key, apikey, authorization, access_token, refresh_token, private_key, ssn, social_security, credit_card, card_number, cvv, client_secret. The value is replaced with [REDACTED] — useful for a stray credential riding along in a tool-call argument or a metadata field, whatever it’s called.
• Secret-shaped text inside a message. Free-text prompt or completion content is scanned for a small set of high-confidence credential shapes and those substrings are replaced in place, without touching the surrounding prose: AWS access key IDs, FleetWarrant Agent API keys and registration secrets, OpenAI API keys, Anthropic API keys, Generic bearer tokens.

Something not working, a question about a request, or you need credentials to get started in the first place — email hello@fleetwarrant.com. Onboarding is hands-on at this stage, so that’s a direct line to the team that set your Tenant up, not a ticket queue.