apache/maka

★ 2,016+148 todayTypeScript

An AI agent workspace that keeps everything running on your own machine, with every action logged

Apache Maka (Incubating) is a local-first AI agent workspace that lets an AI agent read files, run tools, and produce work outputs while keeping data on the user's computer by default. Every model message, tool call, tool result, permission decision, and stop event is written to an append-only log, and the app's sessions, UI, and recovery are all built by replaying that log. It ships as a desktop app (macOS Apple Silicon), a terminal interface, a non-interactive CLI, and an evaluation harness for benchmarking agents.

What it does

What it does: gives an AI agent controlled permission to inspect a project, run tools like Read/Write/Edit/Bash/Glob/Grep, and produce artifacts, instead of just answering chat questions.
How: a single 'Runtime Host' owns sessions, turns, tool execution, and events, writing them to a Runtime Event Log stored locally in SQLite; the app's screens are just views computed from that log, and old tool results can be pruned or compacted without deleting the underlying record.
Where it runs: Desktop app (Electron+React, signed macOS Apple Silicon build only for now), a terminal TUI/CLI, and an Eval mode that runs reproducible benchmark experiments across tasks, repetitions, and subjects.
Data handling: sessions, settings, and credentials stay on the local machine (a plaintext credential vault protected by OS file permissions); the current build does not import older credential or chat-history file formats, so upgrading can leave old conversations unreadable unless backed up.
Status: an early public release under active development, with data formats, CLI commands, and some experimental features still subject to change; Computer Use, Intel Mac, Windows, and Linux support are not yet available (Windows has an unsigned preview).

Why it matters

For people building or evaluating AI coding/task agents, it offers a concrete example of local-first architecture where trust and recoverability come from an append-only execution log rather than a hosted service. It also gives a reusable pattern (log-as-runtime, permission-gated tool calls, reproducible eval harness) that other agent projects can study or adopt.

Terms in this repo

local-first · 데이터와 실행이 클라우드 서버가 아니라 사용자 기기에 기본적으로 저장·처리되는 설계 방식
append-only log · 한번 기록되면 수정하지 않고 계속 뒤에 덧붙이기만 하는 기록 방식
Runtime Host · 세션, 턴, 도구 실행, 이벤트 등 실행 전체를 관리하는 단일 주체
TUI · 터미널 안에서 마우스 없이 조작하는 텍스트 기반 사용자 인터페이스
artifact · 에이전트 작업으로 생성된 파일이나 결과물

Repository description (English)

Apache Maka (Incubating) is a local-first AI agent workspace. Model messages, tool calls, tool results, permission decisions, and termination events are recorded as an append-only log.

Open on GitHub

Coverage

아파치 재단, 로컬 실행 AI 에이전트 도구 '마카' 인큐베이팅

Trending repos

All repos →