freeclaude

Use GLM (Z.AI) as the backend for Claude Code — free, no API key.

A local proxy that wraps chat.z.ai behind an Anthropic-compatible API, so Claude Code can talk to it like it's Claude.

Recommendation: use GLM only. GLM (Z.AI) is the supported, recommended backend. A DeepSeek backend exists for experimentation but is not recommended — slower, less reliable tool calling, flaky file uploads, and no server-side session continuation.

Only tested on macOS.

Prerequisites

You must have Claude Code installed, which requires Node.js / npm.

1. Install Node.js (ships with npm) — nodejs.org or via Homebrew:

   brew install node

2. Install Claude Code:

   npm install -g @anthropic-ai/claude-code

(install.sh will run step 2 for you automatically if claude is not on your $PATH — skip with SKIP_CLAUDE=1. Node/npm must already be installed.)

Install

git clone https://github.com/zerox90x90/freeseek.git
cd freeseek
./install.sh

install.sh handles everything:

• creates .venv and installs Python deps (requires Python 3.11+)
• installs Playwright's Chromium
• drops a freeseek launcher into the first writable $HOME/* directory on your $PATH (skip with SKIP_LAUNCHER=1)
• installs the Claude Code CLI via npm if missing (skip with SKIP_CLAUDE=1)
• runs the one-time Z.AI browser login (required — complete it to finish install)

Run

Backend is GLM (Z.AI) — the only recommended path.

freeseek                              # GLM
MODEL=glm-5.1:search freeseek         # pick a GLM model

Or from the repo directly:

./start.sh             # GLM

DeepSeek (not recommended)

Kept for experimentation only. Expect rough edges.

freeseek deepseek
MODEL=deepseek-reasoner freeseek deepseek
./start.sh deepseek

First run on the DeepSeek backend triggers a one-time browser login for chat.deepseek.com.

License

MIT