ExpressGradient/lazarus

Lazarus

Lazarus is a terminal coding agent with persistent Python tools and carryover. It gives the model a long-lived Python workspace, so the agent can inspect files, edit code, run commands, execute tests, and keep useful state alive while it works.

The project is intentionally small: the CLI loop lives in src/lazarus/cli.py, and the persistent Python worker lives in src/lazarus/python_worker.py.

Features

• Chat with a coding agent from your terminal.
• Run against Kimi/Moonshot, OpenAI, Anthropic, or Google providers through kosong.
• Give the agent a persistent run_python tool for workspace actions.
• Preserve Python variables, imports, helper functions, and notes across tool calls.
• Automatically inject an internal carryover request when token usage gets large, then reset chat history while keeping the Python interpreter alive.

Requirements

• Python 3.12+
• uv
• API credentials for whichever provider you use

Install

Install Lazarus as a uv tool:

uv tool install git+https://github.com/ExpressGradient/lazarus
lazarus

Or run it without installing:

uvx --from git+https://github.com/ExpressGradient/lazarus lazarus

Quit the CLI with:

/quit

Run a single non-interactive job and exit:

lazarus --prompt "summarize this repository"

Providers And Models

Lazarus defaults to the Kimi provider with kimi-k2.6:

lazarus

Choose a provider and model explicitly:

lazarus --provider kimi --model kimi-k2.6
lazarus --provider openai --model gpt-5.2
lazarus --provider anthropic --model claude-sonnet-4.6
lazarus --provider google --model gemini-3-pro

Supported provider aliases:

anthropic
gemini
google
google-genai
kimi
moonshot
openai
openai-legacy
openai-responses

Set credentials using the environment variables expected by the selected kosong provider. For example, Kimi/Moonshot setups commonly use:

export KIMI_AP