Antigma

What is Antigma?

Antigma is the lab behind Ante, an open-source coding agent that lives in your terminal. Ante ships as one small Rust binary with no runtime dependencies, works with more than a dozen model providers, and can run local GGUF models fully offline. It writes code, runs tests, and fixes its own mistakes.

Top Features:

• Single binary: a 15 MB download installs with one curl command and no setup.
• Offline mode: built-in llama.cpp runs GGUF models locally without keys or internet access.
• Any provider: switch between Anthropic, OpenAI, Gemini, Grok, and OpenRouter during a session.

Use Cases:

• Feature building: describe an API in one sentence and get scaffolded, tested code back.
• Private codebases: keep sensitive source code on your machine by using local models.
• Agent coordination: let it drive Claude Code or Codex sessions in shared terminals.

Who Can Use Antigma?

• Software engineers: get a fast terminal assistant that uses far less memory.
• Platform teams: run hundreds of lightweight agent replicas in parallel on modest hardware.
• Privacy-focused developers: code with open-weight models without sending data to the cloud.

Pricing

• Ante CLI (free): open-source under Apache 2.0 and free for commercial use during preview.
• Cloud models (BYOK): pay your chosen provider directly for the tokens each session uses.
• Antix Console (contact sales): team LLM gateway with virtual keys and spend caps.

Pros and Cons

Pros:

• Tiny footprint: uses about seven times less peak memory than Claude Code in benchmarks.
• No lock-in: use your own keys, subscriptions, or local models without an account.
• Public benchmarks: results on Terminal-Bench 2.1 are published with raw runs for auditing.

Cons:

• Beta software: the preview build warns of breaking changes and incomplete features.
• Limited platforms: it runs on macOS and Linux, while Windows users need WSL.
• Terminal first: the main interface is a terminal UI built for keyboard users.

FAQs:

1) Is Ante free to use?

Yes, it is open source, and you only pay the model providers you use.

2) Can it work without internet?

Yes, point it at a local GGUF model and everything runs offline.

3) Which model providers are supported?

Anthropic, OpenAI, Gemini, Grok, OpenRouter, Vertex AI, and any OpenAI-compatible API.

4) How do I install it?

Run one curl install script on macOS or Linux, then type ante.

5) Does it support MCP and subagents?

Yes, along with custom skills, persistent memory, and headless or server modes.