Package Exports

This package does not declare an exports field, so the exports above have been automatically detected and optimized by JSPM instead. If any package subpath is missing, it is recommended to post an issue to the original package (junie-api) to support the "exports" field. If that is not possible, create a JSPM override to customize the exports field for this package.

Readme

Junie API Proxy

[!WARNING] This is a reverse-engineered proxy of JetBrains Junie/Grazie AI API. It is not supported by JetBrains, and may break unexpectedly. Use at your own risk.

[!WARNING] JetBrains Terms of Service: Use of JetBrains AI services is subject to the JetBrains AI Service Terms of Service. Excessive automated or scripted use may violate these terms and could result in suspension of your account or access.

This proxy is intended for personal use with your own JetBrains/Junie subscription. Please use responsibly.

A proxy for the JetBrains Junie/Grazie AI API that exposes it as an OpenAI and Anthropic compatible service. Use your Junie subscription with any tool that supports these APIs, including Claude Code.

Features

OpenAI Compatibility: Exposes Junie AI as an OpenAI-compatible API (/v1/chat/completions, /v1/models)
Anthropic Compatibility: Exposes Junie AI as an Anthropic-compatible API (/v1/messages)
Claude Code Integration: Launch Claude Code with Junie as its backend via --claude-code flag
Multiple Auth Methods: JetBrains OAuth device flow or direct API key
Rate Limiting: Control API usage with --rate-limit and --wait options

Quick Start

npx junie-api

That's it — the first run will guide you through JetBrains OAuth. The server starts on http://localhost:4141.

With Claude Code

npx junie-api --claude-code

With an API key

npx junie-api --auth-token YOUR_API_KEY

Get your key from junie.jetbrains.com/cli.

API Endpoints

Endpoint	Method	Description
`POST /v1/chat/completions`	POST	OpenAI-compatible chat completions
`GET /v1/models`	GET	List available models
`POST /v1/messages`	POST	Anthropic-compatible messages

Claude Code Setup

Use the --claude-code flag, or manually add to .claude/settings.json:

{
  "env": {
    "ANTHROPIC_BASE_URL": "http://localhost:4141",
    "ANTHROPIC_AUTH_TOKEN": "dummy",
    "ANTHROPIC_MODEL": "google/gemini-2.5-pro",
    "ANTHROPIC_SMALL_FAST_MODEL": "google/gemini-2.5-flash",
    "DISABLE_NON_ESSENTIAL_MODEL_CALLS": "1",
    "CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": "1"
  }
}

CLI Reference

npx junie-api                          # Start server (default)
npx junie-api start                    # Same as above
npx junie-api start --port 8080 -v     # Custom port, verbose
npx junie-api start --rate-limit 30 -w # Rate limit: 30s, wait
npx junie-api start --claude-code      # With Claude Code setup
npx junie-api auth                     # Authenticate only

Option	Description	Default	Alias
--port	Port to listen on	4141	-p
--verbose	Enable verbose logging	false	-v
--auth-token	Provide auth token directly	none	-a
--show-token	Show tokens on fetch/refresh	false
--rate-limit	Rate limit in seconds	none	-r
--wait	Wait instead of error on rate limit	false	-w
--claude-code	Generate Claude Code launch command	false	-c

Development

npm install
npm run build       # Build with tsdown
npm test            # Run tests
npm run dev         # Dev with --watch
npm run typecheck   # TypeScript check

Architecture

Two routing paths:

Anthropic Native Passthrough (`/v1/messages` with `claude-*` models)

Forwards requests to the upstream Anthropic-compatible endpoint
Sanitizes unknown fields before forwarding (e.g., context_management)
Streaming: pipes upstream SSE directly to client

Translation Path (`/v1/messages` with non-Claude models, `/v1/chat/completions`)

Translates Anthropic format → OpenAI format → Grazie native protocol
Wraps in { profile, chat: { messages } } for the Grazie gateway
Translates responses back to the requested format

License

MIT

junie-api