Point your editor

Every block below is the string the app's onboarding shows, checked against the source by a test on every build. The endpoint is http://127.0.0.1:11711/v1, the model is rumrunner-auto, and the key is a placeholder — the daemon never checks it; OpenAI clients simply refuse to send without one.

Continue

~/.continue/config.yaml

name: Rumrunner
version: 1.0.0
schema: v1
models:
  - name: Rumrunner
    provider: openai
    model: rumrunner-auto
    apiBase: http://127.0.0.1:11711/v1
    apiKey: rr-local
    roles:
      - chat
      - edit
      - apply
    requestOptions:
      headers:
        X-RR-App: continue

The apiKey is a placeholder. Nothing is sent to OpenAI; the router decides the lane and Rumrunner uses your own keys when a request leaves the Mac. name, version and schema are required — without them Continue reports a config error and lists no model. Run by hand against Continue 2.0.0.

aider

~/.aider.conf.yml

openai-api-base: http://127.0.0.1:11711/v1
openai-api-key: rr-local
model: openai/rumrunner-auto

The same three settings work as OPENAI_API_BASE, OPENAI_API_KEY and --model if you would rather not keep a config file. Run by hand against aider 0.86.2: it connected, streamed, and applied four successive edits.

aider sends the generic OpenAI/Python User-Agent, so on its own its requests look like any other Python client's. An optional second file makes it identify itself, which is what lets the receipt break savings down per app:

~/.aider.model.settings.yml

- name: openai/rumrunner-auto
  extra_params:
    extra_headers:
      X-RR-App: aider

Zed

~/.config/zed/settings.json

{
  "language_models": {
    "openai": {
      "api_url": "http://127.0.0.1:11711/v1",
      "available_models": [
        {
          "name": "rumrunner-auto",
          "display_name": "Rumrunner",
          "max_tokens": 128000
        }
      ]
    }
  }
}

Zed still asks for an OpenAI API key in the assistant panel before it will send anything. Any value does. This snippet has not yet been run against a real Zed; if you do, tell us what happened.

Cline

Cline settings, in VS Code — there is no config file to paste into

API Provider   OpenAI Compatible
Base URL       http://127.0.0.1:11711/v1
API Key        rr-local
Model ID       rumrunner-auto

Cline has no place to set a custom header, so its requests arrive under whatever User-Agent its HTTP client sends and the app reports what it saw rather than guessing. Not yet driven by hand; the settings are what its panel asks for.

Claude Code — not on this build

your shell profile

export ANTHROPIC_BASE_URL=http://127.0.0.1:11711
export ANTHROPIC_AUTH_TOKEN=rr-local

These are the right settings; the route they need — POST /v1/messages, the Anthropic dialect — is the one Rumrunner does not serve yet. Until it does the daemon answers 404 and there is nothing to wait on.

Cursor

Cursor is not on this list and will not be. It routes every request through its own backend, so it cannot reach 127.0.0.1, and even tunnelled the prompt would reach Cursor's servers before it reached this Mac. That is a property of Cursor, not a setting.