Ante by Antigma

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.