OpenAI-Compatible Endpoint for Open WebUI

This playbook exposes a mono-agent agent as an OpenAI-compatible /v1 endpoint so Open WebUI (or any OpenAI client) can list the model, stream responses token-by-token over SSE, and keep multi-turn conversation state per chat.

Who this is for

AI infra engineers fronting the agent with a chat UI.

Goal

Expose the agent as an OpenAI-compatible /v1 endpoint so Open WebUI can stream responses and keep multi-turn conversation state.

Features used

openai-api.chat-completions — /v1/models + /v1/chat/completions with SSE streaming and an optional bearer key.
openai-api.session-headers — session continuity via the X-OpenWebUI-Chat-Id / X-Conversation-Id request headers (only the latest user turn is forwarded per conversation).
runtime.provider-sessions — continuous provider sessions that retain context across requests.

All three are config coverage — no code required.

Configuration

{
  "runtime": {
    "model": "anthropic:claude-sonnet-4-6",
    "session": {
      "mode": "continuous",
      "idleTimeoutMs": 1800000
    }
  },
  "openaiApi": {
    "enabled": true,
    "host": "0.0.0.0",
    "port": 4040,
    "basePath": "/v1",
    "allowNonLoopback": true,
    "modelId": "my-agent"
  }
}

Create an owner-only .env in the invocation folder (chmod 600 .env) and add:

MONO_AGENT_OPENAI_API_KEY=<strong-client-bearer>

The same settings can be supplied via environment variables (MONO_AGENT_*); see Environment variables for the full mapping.

Steps

mono-agent init --model anthropic:claude-sonnet-4-6
Add the openaiApi section; set allowNonLoopback: true for a non-loopback bind, put MONO_AGENT_OPENAI_API_KEY in the owner-only .env, set modelId, and enable continuous session mode under runtime.session.
mono-agent validate, then mono-agent start and confirm the status line reports openaiApi running with its endpoint.
In Open WebUI, add an OpenAI connection pointing at http://host:4040/v1 with the bearer key.
Send two consecutive messages in one Open WebUI chat and confirm continuity (only the latest user turn is forwarded per conversation; prior context comes from the continuous session keyed by the chat id header).
Verify SSE streaming token-by-token in the UI.