gitlawb-opengateway-proxy

name	gitlawb-opengateway-proxy
description	Install, run, and integrate the gitlawb-opengateway-proxy. Use this whenever the user wants free, no-API-key access to Xiaomi MiMo models (mimo-v2.5-pro, mimo-v2-flash, etc.) from any OpenAI-compatible CLI/SDK or agent UI.

gitlawb-opengateway-proxy — install & integration skill

A tiny zero-dependency Node script that proxies the free, no-auth gateway https://opengateway.gitlawb.com/v1/xiaomi-mimo and re-exposes it locally as a plain OpenAI-compatible REST API at http://localhost:19911/v1.

Use this when:

• The user wants to talk to Xiaomi MiMo from a tool that requires an OpenAI base URL (Cline, Continue, Cursor, raw curl, the OpenAI SDK, or any other OpenAI-compatible client).
• The tool insists on a non-empty API key — this proxy lets you use any string (e.g. "none") because the upstream is unauthenticated.
• The tool can't handle the upstream's reasoning_content field — set MERGE_REASONING=1 to fold it into <think>...</think> inside content.

1. Requirements

• Node.js ≥ 18 (uses the global fetch). Check with node -v.
• Port 19911 free (or pick any other via PORT=...).
• Outbound HTTPS to opengateway.gitlawb.com.

2. Install

The proxy is a single file. Two equivalent install paths:

A. Run directly without installing (preferred for one-off testing)

cd /path/to/gitlawb-opengateway-proxy
node server.mjs

B. Install as a global CLI

cd /path/to/gitlawb-opengateway-proxy
npm install -g .
# now usable from anywhere
gitlawb-opengateway-proxy

C. Background daemon via systemd (Linux server)

sudo tee /etc/systemd/system/gitlawb-opengateway-proxy.service >/dev/null <<'UNIT'
[Unit]
Description=Gitlawb Opengateway proxy (Xiaomi MiMo, no-auth)
After=network-online.target

[Service]
Type=simple
ExecStart=/usr/bin/node /opt/gitlawb-opengateway-proxy/server.mjs
Environment=PORT=19911
Restart=on-failure
User=nobody

[Install]
WantedBy=multi-user.target
UNIT

sudo mkdir -p /opt/gitlawb-opengateway-proxy
sudo cp server.mjs /opt/gitlawb-opengateway-proxy/
sudo systemctl daemon-reload
sudo systemctl enable --now gitlawb-opengateway-proxy
sudo systemctl status gitlawb-opengateway-proxy

3. Verify

# health
curl -s http://localhost:19911/health
# -> {"ok":true,"upstream":"https://opengateway.gitlawb.com/v1/xiaomi-mimo",...}

# models
curl -s http://localhost:19911/v1/models | head -c 400

# non-streaming chat completion
curl -s http://localhost:19911/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"mimo-v2.5-pro","messages":[{"role":"user","content":"Say hi"}],"max_tokens":20}'

# streaming chat completion (reasoning_content is preserved)
curl -sN http://localhost:19911/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"mimo-v2.5-pro","messages":[{"role":"user","content":"2+2?"}],"stream":true,"max_tokens":80}'

If curl returns valid JSON / SSE, the proxy is good.

4. Environment variables

Var	Default	Purpose
PORT	19911	TCP port to listen on
UPSTREAM_BASE	https://opengateway.gitlawb.com/v1/xiaomi-mimo	Override the gateway sub-route (e.g. another vendor on opengateway)
MERGE_REASONING	0	If 1, fold reasoning_content into content as <think>...</think> blocks. Turn on for clients that don't render reasoning_content.
DROP_KEEPALIVE	1	Drop the upstream's empty delta:{} keepalive chunks. Set to 0 if your client relies on them.
DEBUG	0	Log every request to stderr

5. Hook into any OpenAI-compatible client

In any client that supports an OpenAI-compatible / custom provider, set:

Field	Value
Provider type	OpenAI-compatible
Base URL	http://localhost:19911/v1
API Key	none (or any non-empty string — required only because most UIs demand one)
Default model	mimo-v2.5-pro (or any other mimo-* listed by /v1/models)

If the client is on a different host, point it at http://<host>:19911/v1 and make sure port 19911 is reachable / firewalled appropriately.

OpenAI SDK example

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "http://localhost:19911/v1",
  apiKey: "none",
});

const res = await client.chat.completions.create({
  model: "mimo-v2.5-pro",
  messages: [{ role: "user", content: "Hello" }],
});
console.log(res.choices[0].message.content);

6. Reasoning / "thinking" output — make sure it never errors

The upstream returns chain-of-thought as delta.reasoning_content (DeepSeek / MiMo convention), not inside content. This proxy passes it through unchanged by default, which works for:

• OpenAI SDK ≥ 1.50 (accepts unknown fields gracefully)
• Anthropic-compat shims that ignore unknown OpenAI fields
• Most modern OpenAI-compatible routers

If your client errors on reasoning_content (rare — usually old, strict parsers), start the proxy with MERGE_REASONING=1. The reasoning text will then be folded into content as a <think>...</think> block, which every markdown renderer handles.

MERGE_REASONING=1 node server.mjs

7. Troubleshooting

Symptom	Fix
400 Param Incorrect from upstream	The /v1/xiaomi-mimo route only accepts mimo-* models. Set UPSTREAM_BASE to a different sub-route if you want a different vendor, or stick to mimo-v2.5-pro etc.
Client crashes on empty delta:{} chunks	Default DROP_KEEPALIVE=1 removes them — make sure you didn't set DROP_KEEPALIVE=0.
EADDRINUSE :::19911	Pick another port: PORT=29911 node server.mjs.
Streaming hangs in browser fetch	Ensure your reverse proxy / CDN doesn't buffer SSE. The proxy already sets X-Accel-Buffering: no.
No reasoning visible in UI	Either your client doesn't render reasoning_content — set MERGE_REASONING=1 — or the model wasn't in reasoning mode (use mimo-v2.5-pro).
model is required	Make sure the POST body has "model": "mimo-v2.5-pro".

8. Update / uninstall

# update: just `git pull` and restart, no rebuild
sudo systemctl restart gitlawb-opengateway-proxy   # if installed as a service

# uninstall (systemd)
sudo systemctl disable --now gitlawb-opengateway-proxy
sudo rm /etc/systemd/system/gitlawb-opengateway-proxy.service /opt/gitlawb-opengateway-proxy/server.mjs

# uninstall (global npm)
npm uninstall -g gitlawb-opengateway-proxy