zerostack  by gi-dellav

Efficient Rust coding agent for streamlined development

Created 2 months ago
1,443 stars

Top 27.6% on SourcePulse

GitHubView on GitHub
Project Summary

Minimalistic coding agent written in Rust, optimized for memory footprint and performance. It targets engineers and power users seeking a lean yet powerful AI assistant, offering extensive LLM provider support and advanced features like ACP integration and Git worktree management, significantly outperforming larger JavaScript-based alternatives in resource utilization.

How It Works

Written in Rust for speed and low resource consumption, Zerostack supports multiple LLM providers (OpenRouter, OpenAI, Anthropic, Gemini, Ollama, custom) and integrates standard coding agent tools. Key innovations include a granular permission system with configurable modes and per-tool patterns, robust session management, and an interactive terminal UI with markdown rendering. The agent dynamically switches between specialized system prompts (e.g., code, plan, review, debug) to tailor its behavior.

Quick Start & Requirements

Install via Cargo: cargo install zerostack. For ACP editor integration, use cargo install zerostack --features acp. All features require cargo install zerostack --features "acp,loop,git-worktree,mcp". Pre-built binaries are available on GitHub Releases. Prerequisites: Rust/Cargo, Git. Optional: bubblewrap for sandbox mode. API keys (e.g., OPENROUTER_API_KEY) are needed. See CONFIG.md for details.

Highlighted Details

  • Performance: ~7k LoC, 8.9MB binary, ~8MB RAM idle, ~12MB RAM active, ~1.5% CPU active.
  • Multi-Provider Support: Integrates OpenRouter, OpenAI-compatible, Anthropic, Gemini, Ollama, and custom providers.
  • Advanced Features: Integrated Exa search, experimental loop capabilities, experimental Git worktree integration, gated ACP support for editors.
  • Prompting & Permissions: Built-in system prompts (code, plan, review, etc.) and a four-mode permission system (restrictive, standard, accept-all, yolo) with per-tool glob patterns and doom-loop detection.

Maintenance & Community

The provided README does not detail maintainers, community channels, or sponsorships.

Licensing & Compatibility

Licensed under GPL-3.0-only, a strong copyleft license. Windows support is explicitly noted as untested.

Limitations & Caveats

Loop system and Git worktrees integration are experimental. Windows compatibility is not guaranteed. The GPL-3.0-only license requires careful consideration for commercial adoption or integration into closed-source projects.

Health Check
Last Commit

1 day ago

Responsiveness

Inactive

Pull Requests (30d)
61
Issues (30d)
15
Star History
202 stars in the last 30 days

Explore Similar Projects

Starred by Patrick von Platen Patrick von Platen(Author of Hugging Face Diffusers; Research Engineer at Mistral), Simon Willison Simon Willison(Coauthor of Django), and
1 more.

mistral-vibe by mistralai

0.7%
5k
Conversational CLI coding agent for project interaction
Created 7 months ago
Updated 1 day ago
Starred by Sebastian Raschka Sebastian Raschka(Author of "Build a Large Language Model (From Scratch)"), Jeff Hammerbacher Jeff Hammerbacher(Cofounder of Cloudera), and
9 more.

qwen-code by QwenLM

0.7%
26k
AI coding agent for complex codebases
Created 1 year ago
Updated 17 hours ago
Feedback? Help us improve.