AntigmaLabs/ante-preview

Ghost in your shell. Ante is a self-contained agent harness with a highly optimized core. It works like Claude Code or Codex, with none of their dependencies or model constraints.

What it solves

Ante is a high-performance, self-contained coding agent designed to live in the terminal. It eliminates the heavy runtime dependencies (like Node.js or Python) and vendor lock-in common in other AI agents, providing a lightweight alternative that can be run locally, offline, or via various API providers.

How it works

Built in native Rust, Ante is distributed as a single ~15MB binary. It embeds essential tools like Grep and git directly into the same process to minimize resource leakage and overhead. It features a client-daemon architecture that allows it to be used as an interactive TUI, a headless CLI, or a long-lived server. Additionally, it includes a built-in inference engine based on llama.cpp to run GGUF models natively offline.

Who it’s for

Developers who want a fast, resource-efficient coding assistant that can be run entirely offline or across multiple LLM providers without requiring an account or external runtime environments.

Highlights

  • Zero Dependencies: A single Rust binary with no external runtime requirements.
  • Native Offline Mode: Built-in inference engine for running GGUF models without an internet connection.
  • Multi-Provider Support: Compatible with 12+ providers including Anthropic, OpenAI, Gemini, and Grok.
  • Extreme Efficiency: Uses significantly less peak memory, CPU, and disk I/O compared to similar tools like Claude Code.
  • Flexible Deployment: Can be run as a TUI, headless CLI, or a server (ante serve) or a gateway for Slack/Discord.
  • Multi-Agent Orchestration: Ability to spawn sub-agents to coordinate complex tasks.