qwen-kolibri-rust-gateway

Fast local OpenAI-compatible Rust gateway for Qwen Code CLI, OpenAI SDK clients, curl, IDE agents, and future Kolibri API deployment.

Architecture

OpenAI SDK / Qwen Code CLI / curl
→ http://127.0.0.1:8787/v1
→ Kolibri Rust Gateway
→ upstream OpenAI-compatible provider

First upstream target:

https://kolibriai.ru/v1

Future upstream target:

https://api.kolibriai.ru/v1

Clients only change baseURL.

What this gateway does

• Local OpenAI-compatible /v1/chat/completions gateway.
• Passthrough to OpenAI-compatible upstream.
• Stream passthrough for SSE.
• Tool calls passthrough.
• exit_plan_mode passthrough.
• Model suffix routing:
  • qwen3.6-plus#project-a → upstream model qwen3.6-plus, session key project-a.
  • qwen/qwen3.6-plus#project-a → upstream model qwen3.6-plus, session key project-a.
• Optional enable_thinking injection.
• Local/Upstream API key separation.
• Secret masking in debug output.
• qwen-local wrapper over the original Qwen Code CLI.

What it intentionally does not do

• No direct chat.qwen.ai/api/v2 browser reverse-engineering.
• No custom agent loop.
• No local tool execution.
• No rewrite of Qwen CLI.
• No prompt-based fake JSON tool call generation.

Setup

cp .env.example .env
nano .env

Set:

UPSTREAM_API_KEY=your_aikit_or_upstream_key
LOCAL_API_KEY=local-proxy-key-2026

Build and run:

cargo check
cargo run

Health:

curl -s http://127.0.0.1:8787/health | python3 -m json.tool

Models:

curl -s http://127.0.0.1:8787/v1/models \
  -H "Authorization: Bearer local-proxy-key-2026" \
  | python3 -m json.tool

Chat non-stream:

curl -s http://127.0.0.1:8787/v1/chat/completions \
  -H "Authorization: Bearer local-proxy-key-2026" \
  -H "Content-Type: application/json" \
  -d '{"model":"qwen3.6-plus#project-a","stream":false,"messages":[{"role":"user","content":"Ответь одним словом: работает?"}]}' \
  | python3 -m json.tool

Chat stream:

curl -N http://127.0.0.1:8787/v1/chat/completions \
  -H "Authorization: Bearer local-proxy-key-2026" \
  -H "Content-Type: application/json" \
  -d '{"model":"qwen3.6-plus#project-a","stream":true,"messages":[{"role":"user","content":"Напиши три слова: тест bridge работает"}]}'

OpenAI SDK usage

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.KOLIBRI_API_KEY || "local-proxy-key-2026",
  baseURL: process.env.KOLIBRI_BASE_URL || "http://127.0.0.1:8787/v1",
});

const completion = await client.chat.completions.create({
  model: "qwen3.6-plus#my-project",
  messages: [{ role: "user", content: "Ответь одним словом: работает?" }],
});

console.log(completion.choices[0].message.content);

Global Installation

Install globally via npm to use qwen-kolibri from any directory:

npm install -g qwen-kolibri-rust-gateway

After installation, run from anywhere:

qwen-kolibri [options]

Custom KOLIBRI Banner

The wrapper displays a colorful ASCII banner before launching Qwen Code:

█▀█▀█▀█  ██  ██ █▀█▀█▀█ █▀█▀    ██
  ██╔═══██╗██║    ██║██╔════╝████╗  ██║  │ >_ KOLIBRI Gateway                                  │
  ██║   ██║██║ █╗ ██║█████╗  ██╔██╗ ██║  │                                                          │
  ██║▄▄ ██║██║███╗██║██╔══╝  ██║╚██╗██║  │ API Key | qwen3.6-plus (/model to change)                │
  ╚██████╔╝╚███╔███╔╝███████╗██║ ╚████║  │ ~/your/project                    │
   ╚══▀▀═╝  ╚══╝╚══╝ ╚══════╝╚═╝  ╚═══╝  └──────────────────────────────────────────────────────────┘

Colors: red (K, L, B), green (O, I), yellow (I, R) via ANSI escape codes.

Disable banner: use --no-banner flag:

qwen-kolibri --no-banner [other args]

The wrapper:

• loads .env;
• starts the gateway if not already running;
• creates project-specific model suffix;
• creates isolated Qwen settings under .qwen-kolibri/settings.json;
• runs original qwen directly without script | python3.

Custom KOLIBRI Banner

The bin/qwen-local wrapper displays a colorful KOLIBRI banner before launching Qwen Code:

█▀█▀█▀█  ██  ██ █▀█▀█▀█ █▀█▀    ██
  ██╔═══██╗██║    ██║██╔════╝████╗  ██║  │ >_ KOLIBRI Gateway                                  │
  ██║   ██║██║ █╗ ██║█████╗  ██╔██╗ ██║  │                                                          │
  ██║▄▄ ██║██║███╗██║██╔══╝  ██║╚██╗██║  │ API Key | qwen3.6-plus (/model to change)                │
  ╚██████╔╝╚███╔███╔╝███████╗██║ ╚████║  │ ~/your/project                    │
   ╚══▀▀═╝  ╚══╝╚══╝ ╚══════╝╚═╝  ╚═══╝  └──────────────────────────────────────────────────────────┘

Colors: red (K, L, B), green (O, I), yellow (I, R) via ANSI escape codes.

Disable banner: use --no-banner flag:

qwen-local --no-banner [other args]

Benchmark

export AIKIT_API_KEY="your_key"
./scripts/bench_aikit_vs_gateway.sh

Output:

local-data/bench/<timestamp>/

Deployment path

Local:

OPENAI_BASE_URL=http://127.0.0.1:8787/v1

Future server:

OPENAI_BASE_URL=https://api.kolibriai.ru/v1

Client code does not change.