Skip to main content
Akio is under active development. The project is not yet fully functional and may not work as expected out of the box. If you’d like to contribute, feel free to open a pull request. Found a bug, vulnerability, or something unusual? Please open an issue. Akio is primarily developed for Linux and macOS. It probably won’t work as intended on Windows, even if it compiles.

​
What is Akio?

Akio is a plug-and-play autonomous AI agent with embedded model inference, written in Rust. No Anthropic. No OpenAI. No Ollama. The inference runs directly inside the binary using llama.cpp via FFI. Never depend on a model provider or Google a command again. It combines the power of large language models with real-world capabilities through built-in tools and MCP server support.

Quickstart

Install Akio, pull a model, and start your first session.

CLI Reference

Full reference for all akio commands and flags.

Built-in Tools

Shell, read, write, glob, and websearch — out of the box.

MCP Servers

Extend Akio with any MCP-compatible tool server.

​
Key features

  • In-binary inference — ships with llama.cpp compiled in; no external runtime needed
  • Default tools — shell execution, file read/write, glob search, and web search
  • MCP support — connect any Model Context Protocol server for additional capabilities
  • GPU acceleration — offload layers to GPU via --ngl
  • Fully offline — once a model is pulled, no internet connection required

​
How it works

akio pull ggml-org/Qwen3-8B-GGUF   # Download a GGUF model from HuggingFace
akio run -m Qwen3-8B-Q4_K_M.gguf   # Start an interactive session
Akio loads the model into memory, exposes a set of tools to the LLM, and runs an autonomous agentic loop — the model reasons, calls tools, observes results, and continues until the task is complete.

​
Source