AI Gateway user manual

Internally, Node Neo uses the embedded Morpheus proxy-router SDK for inference. The AI Gateway is a small HTTP server in front of that SDK that speaks the same JSON shape as OpenAI's API: /v1/chat/completions, /v1/models, and related routes. External apps point their “base URL” at Node Neo instead of OpenAI.

The gateway does not replace the Morpheus network or run models offline — it routes requests through your wallet's on-chain sessions on Base, just like chat in the UI. API-driven chats can appear in the app as conversations sourced from the API.

Platform: In the shipping app, AI Gateway controls appear under Expert Mode on desktop platforms (macOS today; Linux and Windows builds follow the same Expert Mode layout when available). Mobile builds focus on consumer chat and do not expose this server.

1. Wallet ready. You need a Node Neo wallet with MOR (session stake) and ETH on Base (gas), same as in-app chat. See Quick start or New to crypto?.
2. Keep Node Neo running. The gateway process lives inside the app; when you quit Node Neo, the HTTP server stops.
3. Know the difference from Developer API. Expert Mode also offers a separate Developer API (Swagger, different port, HTTP Basic auth). That is not the OpenAI-compatible gateway. This page is only about the AI Gateway.

1. Open Node Neo on your desktop.
2. Go to Settings → Expert Mode.
3. Scroll to the AI Gateway section.
4. Choose listen scope (see Localhost vs LAN below).
5. Set the TCP port. Default is 8083; change it only if something else already uses that port.
6. Tap Start Gateway. The screen shows the address the server bound to (for example 127.0.0.1:8083 or 0.0.0.0:8083).

The Expert screen also shows a base URL hint for OpenAI-style clients (for example http://127.0.0.1:8083/v1) and reminds you to use an API key from the section below.

1. In Expert Mode, same section as the gateway, open API Keys.
2. Tap Generate Key. Optionally label it (for example “Zed”).
3. Copy the full sk-... secret immediately; the app stores only a hash and cannot show it again.
4. Pass the key as Authorization: Bearer <your-key> on every gateway request except GET /health.
5. To revoke access, delete the key in the list; requests using that token fail right away.

Zed documents OpenAI-compatible providers under LLM Providers: you add a provider with a custom api_url and list the models Zed should show in the Agent panel. Official reference: zed.dev/docs/ai/llm-providers (see the “OpenAI API Compatible” section).

1. Start the gateway in Local only mode so Zed on the same Mac talks to 127.0.0.1.
2. Pick your port (default 8083). Base URL for OpenAI-style clients is http://127.0.0.1:<port>/v1 — include the /v1 suffix.
3. Call GET /v1/models (with curl above) and note the exact model id strings you want in Zed.
4. In Zed, open Agent settings (agent: open settings), use Add Provider, choose an OpenAI-compatible profile, and paste the base URL and API key. Alternatively, merge a snippet like the following into your Zed settings.json — adjust provider label, model names, and URL to match your machine:

{
  "language_models": {
    "openai_compatible": {
      "Node Neo": {
        "api_url": "http://127.0.0.1:8083/v1",
        "available_models": [
          {
            "name": "your-model-id-from-/v1/models",
            "display_name": "Morpheus (Node Neo)"
          }
        ]
      }
    }
  }
}

Zed expects API keys via the Agent UI or environment variables derived from the provider name; keys are not embedded in the JSON file. Follow the current Zed docs for the exact variable name pattern and keychain behavior.

For Zed on another machine, run Node Neo in Network mode and point api_url at http://<your-lan-ip>:<port>/v1, still using Bearer auth.

Some IDE features that rewrite OpenAI base URLs route traffic through vendor infrastructure, which blocks localhost. For Cursor, the Node Neo repo ships an MCP (Model Context Protocol) server that talks to your local gateway over HTTP from a stdio subprocess instead.

Advanced setup lives in the repository (mcp-server/) and requires building the Node package, pointing NODENEO_GATEWAY_URL at your running gateway (must match your chosen port), and setting NODENEO_API_KEY. See the project's llms.txt and MCP README when you are ready; details change with releases.

• Connection refused. Confirm Node Neo is open and Start Gateway was pressed; verify port and that you use http:// not https://.
• 401 from API routes. Bearer token missing, wrong key, or revoked key — generate a fresh secret if unsure.
• Works on localhost but not from another device. Switch to Network bind mode, confirm LAN IP, and check macOS firewall inbound rules for that TCP port.
• Model unknown errors. Use exact ids from GET /v1/models; Morpheus rotates advertised models over time.

More context: Support, Deep dive, or open an issue on GitHub.