Octomind – open-source AI harness that saves your tokens, static rust

Octomind is a CLI-first AI coding agent runtime that allows LLMs real access to a developer's repository through MCP tools. It can read and write files, run shells, search code, delegate to sub-agents, and work in various environments such as the terminal, CI, or as a daemon. Octomind can also be installed as a background process or piped through stdin.

The platform offers several installation options, including Homebrew on macOS, environment variables for custom model access, and a built-in session engine that adapts to different workflows. The default interface is a plain line prompt, which supports markdown rendering and works over SSH, in tmux, and in CI logs. When running a specialist for the first time, the agent may install its tools and ask for credentials.

Octomind provides a default role, assistant:concierge, but users can create custom agents by using their own taps or the community-provided ones. Guardrails, a feature that enforces deterministic runtime behavior, can be configured using a .agents/guardrails.toml file in the repository. The platform supports machine-readable output through the --schema option and structured output when a model supports structured-output.

The installation process varies based on the operating system. On macOS and Linux, it installs to ~/.local/bin/, while on Windows, it installs to %LOCALAPPDATA%\octomind\bin. Users can override the installation directory using the --install-dir or OCTOMIND_INSTALL_DIR environment variable. The source code is available for building from source, requiring protoc for building and a set of OS-specific prerequisites.