Direct Isaac API

POST /v1/chat/completions exposes raw Pokee Isaac without the agent loop, tools, or persistent session state. It accepts the OpenAI Chat Completions request shape and supports both JSON and Server-Sent Events responses.

Production security: This is a direct API key and endpoint. The bearer currently grants direct access to your tenant gateway, so do not embed it in browser, mobile, desktop, or other distributed client code. For production, we recommend calling Pokee through your own server-side proxy or backend-for-frontend, with the key stored in a secret manager. Your proxy should authenticate your users, enforce application-specific limits, and forward only the request fields your application needs. Pokee will add API-key minting for production integrations so you can issue separate credentials instead of distributing the primary tenant key.

Endpoint and authentication

The route is hosted on your existing dedicated tenant endpoint:

POST https://your-tenant.enterprise.pokee.ai/v1/chat/completions
Authorization: Bearer pk_<tenant>_<token>

The direct Isaac route is enabled per tenant. A tenant where it has not been enabled returns 404. The route currently accepts only model: "pokee-isaac".

Quickstart

export POKEE_API="https://your-tenant.enterprise.pokee.ai"
export POKEE_KEY="pk_<tenant>_<token>"

curl -X POST "$POKEE_API/v1/chat/completions" \
  -H "Authorization: Bearer $POKEE_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "pokee-isaac",
    "messages": [
      {"role": "user", "content": "Reply with exactly: hello from Isaac"}
    ],
    "max_tokens": 64,
    "temperature": 0
  }'

The response uses the standard Chat Completions shape and includes authoritative token usage:

{
  "id": "chatcmpl-...",
  "object": "chat.completion",
  "model": "pokee-isaac",
  "choices": [
    {
      "index": 0,
      "message": {"role": "assistant", "content": "hello from Isaac"},
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 18,
    "completion_tokens": 4,
    "total_tokens": 22
  }
}

Streaming

Set "stream": true to receive OpenAI-compatible SSE chunks. The terminal usage chunk is consumed and billed even if the external client disconnects early.

curl -N -X POST "$POKEE_API/v1/chat/completions" \
  -H "Authorization: Bearer $POKEE_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "pokee-isaac",
    "stream": true,
    "messages": [{"role": "user", "content": "Explain compound interest."}]
  }'

Context routing

The gateway automatically routes the raw request to the appropriate Isaac context pool. No endpoint or model change is required:

Pool selection uses a request-size estimate. Billing always uses the authoritative usage reported by Isaac, not the routing estimate.

Billing and limits

Usage is attributed to the authenticated caller and debited from the corresponding tenant credit ledger. GET /v1/usage reports direct calls under the independent isaac_passthrough rate-limit bucket. This keeps raw completion traffic separate from session turns and /v1/responses traffic.