llmux

Scriptable LLM multiplexer — unified interface to Claude and OpenAI via subscription OAuth (no API keys needed).

Getting tokens

Claude

llmux reuses the token from an existing Claude Code (CLI) session.

1. Install Claude Code: npm install -g @anthropic-ai/claude-code
2. Log in: claude login
3. Done — llmux reads ~/.claude/.credentials.json automatically.

Or set the env var manually:

export CLAUDE_OAUTH_TOKEN=your_token_here

The token is at .claudeAiOauth.accessToken inside ~/.claude/.credentials.json.

OpenAI (Codex)

llmux reuses the token from an existing Codex CLI session.

1. Install Codex: npm install -g @openai/codex
2. Log in: codex login
3. Done — llmux reads ~/.codex/auth.json automatically.

Or set env vars manually:

export CODEX_OAUTH_TOKEN=your_access_token
export CODEX_ACCOUNT_ID=your_account_id

Both values are in ~/.codex/auth.json under .tokens.access_token and .tokens.account_id.

Google (Antigravity CLI)

llmux reuses the token from an existing Antigravity CLI (agy) session.

1. Install: curl -fsSL https://antigravity.google/cli/install.sh | bash
2. Log in: agy (follows OAuth flow on first run)
3. Done — llmux reads ~/.gemini/antigravity-cli/antigravity-oauth-token automatically.

Or set the env var manually:

export AGY_OAUTH_TOKEN=your_access_token_here

The access token is at .token.access_token inside ~/.gemini/antigravity-cli/antigravity-oauth-token. Token expires after 1 hour — re-run agy to refresh.

Quick start

from llmux import Agent, Claude, GPT, Gemini

# Claude (default)
agent = Agent(Claude.sonnet46)
print(agent("Hello!"))

# OpenAI
agent = Agent(GPT.gpt55)
print(agent("Hello!"))

# Gemini (via Antigravity CLI)
agent = Agent(Gemini.flash25)
print(agent("Hello!"))

# Streaming
with Agent(Claude.sonnet46) as agent:
    for chunk in agent.stream("Tell me a joke"):
        print(chunk, end="", flush=True)

Install

pip install llmux

From git:

pip install git+ssh://git@gitlab.com/grisus/llmux-sdk.git

Locally (development):

pip install -e ".[dev]"   # editable + pytest
pip install -e .           # runtime only

Run tests

pytest