Ante is a terminal coding agent from Antigma Labs, shipped as a single Rust executable with no runtime dependencies. It works in the same role as Claude Code or Codex but is not tied to one model vendor, and its core can also serve as the base for a custom harness. The project is a beta preview for macOS and Linux.
The project treats the harness and the model as separate concerns. A settings profile defines the whole agent (system prompt, tool set, skills, memory), so the same binary runs as a full assistant or a minimal agent. Curated profiles include pi, which strips Ante to four tools and a short replacement prompt, and plan, a read-only planning agent. Project settings layer tool filters and reasoning effort over user preferences but cannot change the model, add MCP servers, or widen permissions.
Ante is evaluated continuously on Terminal-Bench 2.1 across model families, with each result pinned to a downloadable build. The latest full run scored 83.9% with the open-weight DeepSeek V4.1 Flash. Grep and git are embedded in the binary, and across 20 parallel tasks Ante uses about 7 times less peak memory, 9 times less average CPU, and 5 times less disk I/O than Claude Code.
Local inference runs through a managed, pinned llama.cpp, so a GGUF model runs the whole loop offline with no API key or account. Hosted providers come as built-in presets, and a catalog file adds any other proxy, gateway, or OpenAI-compatible endpoint.
| Provider | Example models |
|---|---|
| Anthropic | Claude Sonnet 5.5, Opus 5.5, Fable 5.1 (API key or subscription) |
| OpenAI | GPT-6 Astra, GPT-6.1 Sol, GPT-6 Luna (API key or ChatGPT/Codex subscription) |
| Google Gemini | Gemini 3.x family (Gemini API or Vertex AI) |
| Grok (xAI) | Grok 4.7 |
| DeepSeek | DeepSeek V4.1 Flash |
| Open Router | Any Open Router model |
| Local (GGUF) | Any GGUF model through llama.cpp |
Features
- Terminal orchestration: the /term command opens other agent CLIs (Ante, Claude Code, Codex) in shared terminals that Ante drives and you can step into, acting as persistent subagents
- Run modes: an interactive TUI, a headless one-shot mode for scripts and CI, a server mode for editor plugins, and a gateway that runs Ante as a Slack or Discord bot
- Named profiles: one settings file defines the system prompt, tools, skills, and memory, and a profile swaps it per run
- Offline inference: a managed llama.cpp runs GGUF models on your machine with no internet
- Provider switching: change provider, model, and reasoning effort mid-session, with API key or subscription login
- Custom providers: a catalog entry adds any endpoint with one of four API dialects and a bearer, header, or query auth style
- Editor support: an Agent Client Protocol server drives Ante from editors such as Zed and JetBrains IDEs
- Rust SDK: an async SDK and a wire protocol control Ante over stdio or Unix domain sockets
- Session control: interrupt and steer the agent, add files with @ mentions, resume saved sessions, and fork conversations
