# Ante by Antigma

> Ante is a self-contained, local-first AI coding agent built in Rust that runs entirely in your terminal with zero external dependencies and optional full offline mode.

Ante is an autonomous coding agent developed by Antigma Labs, built from scratch in native Rust as a single self-contained binary with no external runtime dependencies. It runs directly in the terminal and supports both cloud-based AI providers and fully offline local inference via a built-in llama.cpp integration. Antigma also open-sourced a companion project, nanochat-rs, a tiny GPT-style cognitive core written in pure Rust inspired by Andrej Karpathy's nanochat.

## What It Is

Ante is a terminal-native AI coding agent designed around the concept of "cellular-native agents" — lightweight, massively replicable, and self-healing. The agent reads a codebase, accepts a task, and produces working output autonomously. It is positioned as a resource-efficient alternative to tools like Claude Code, with Antigma claiming 7× less peak memory, 9× less average CPU usage, and 5× less total disk I/O in benchmarks run with 20 parallel tasks using the same model and prompts.

## Architecture and Engineering Principles

Ante is built on four stated engineering principles: lightweight, reliable, closed-loop, and minimal cognitive load. The binary weighs approximately 15 MB and carries zero runtime dependencies. Its closed-loop design uses declarative intent with automatic reconciliation, meaning individual agent instances are treated as expendable while the overall system persists. The offline mode installs a local inference engine on demand via a single `/offline-mode` command, requiring no API keys or internet connection after setup.

## Setup Path

Getting started with Ante requires a single curl command:

- `curl -fsSL https://ante.run/install.sh | bash`
- No accounts, no configuration files, no setup guides required for basic use
- Antix (the companion web portal) provides account management, API key handling, and a dashboard
- Supports Anthropic, OpenAI, Gemini, Grok, Open Router, and more via bring-your-own-key
- Local model support via built-in llama.cpp; model, context window, and thinking mode are all configurable

## nanochat-rs: The Open-Source Cognitive Core

Alongside Ante, Antigma open-sourced nanochat-rs — a Rust implementation of karpathy/nanochat built on Hugging Face's Candle tensor library. It features a centralized model loader resilient to tensor name changes, a production-ready tokenizer compatible with tiktoken `.pkl` configs, and a ChatGPT-style local web UI. Pre-trained weights (32-layer and 20-layer variants) are available on Hugging Face under the Antigma organization. The repository is dual-licensed MIT/Apache-2.0 and had 68 stars as of its last recorded update.

## Team Background

Antigma Labs was founded by a team that, according to the About page, includes former Principal and Staff Engineers from FAANG companies, a founding engineer from a unicorn startup, a top quantitative trader from a hedge fund, an ICPC medalist, and computer science researchers. The company name references Alan Turing's anti-Enigma work. The flagship product Ante was formally introduced in a March 2026 blog post.

## Update: Ante Launch and Recent Blog Activity

Ante was publicly introduced on March 31, 2026, with a blog post titled "Introduce Ante: Self-Contained Agent That Self-Organize." A follow-up post on April 23, 2026, covered offline coding models reaching a "console moment." Earlier posts include a March 2026 piece on achieving #1 on Terminal Bench and a mid-2025 post on model abliteration. The nanochat-rs repository was last pushed in February 2026 and last updated in May 2026, indicating active development across both projects.

## Features
- Single self-contained Rust binary (~15 MB, zero runtime dependencies)
- Built-in offline mode via llama.cpp integration
- Bring-your-own API key support (Anthropic, OpenAI, Gemini, Grok, Open Router)
- Terminal-native agent that reads codebases and produces working output
- Cellular-native agent architecture for parallel task execution
- Antix portal for account, API key, and dashboard management
- Configurable model, context window, and thinking mode
- 7× less peak memory vs Claude Code (vendor benchmark)
- 9× less average CPU vs Claude Code (vendor benchmark)
- Closed-loop declarative intent with automatic reconciliation

## Integrations
Anthropic Claude, OpenAI, Google Gemini, Grok, Open Router, llama.cpp (local inference), Hugging Face

## Platforms
MACOS, WEB, API, CLI

## Pricing
Freemium — Free tier available with paid upgrades

## Links
- Website: https://antigma.ai
- Documentation: https://docs.antigma.ai/
- Repository: https://github.com/AntigmaLabs/nanochat-rs
- EveryDev.ai: https://www.everydev.ai/tools/ante-antigma