maka

Why we included this project

Maka is for people who want an agent whose work they can go back and audit, not just take at face value. Runs are recorded in an append-only log that stays on your machine by default, capturing model messages, tool calls, tool results, permission decisions, and termination events, and the interface is simply a projection over that log. You can prune or compact tool results to shape what the next inference sees without destroying the evidence, and connect the model to a cloud API, a local model, or a compatible gateway. It ships as a desktop app, a terminal UI, and a non-interactive CLI, plus an eval harness for reproducible benchmark runs. One caveat: it is still under active development, so the early macOS Apple Silicon build may still shift its formats and commands.