# Apache Maka

> Apache-2.0 local-first agent workspace that logs every step

**Category:** AI Agents  
**Pricing:** open-source  
**URL:** https://maka.apache.org/en/  
**GitHub:** https://github.com/apache/maka  
**Added:** 2026-10-17

---

## The Hook

Maka is a coding-agent workspace (desktop app, TUI and CLI) where every model message, tool call, permission decision and result is written to an append-only event log, so you can replay, resume and audit exactly what the agent did on your machine.

## What's Good

Apache-2.0 (apache/maka), an Apache Incubator podling since 2026-08-13, very active (commits daily, nightly builds). Desktop, TUI, CLI and the eval runner are thin clients of one Runtime Host, and the UI, the next prompt and crash recovery are all built from the same log, so an interrupted turn can be resumed with /resume. Bring your own model: a cloud API, a local model or a compatible gateway. It has MCP client support, a --graph mode that splits work across isolated git worktrees, and the team publishes per-task Terminal-Bench runs against other harnesses in docs/eval instead of only claiming scores. Sessions and settings stay local. ★~5.8k.

## The Catch

Early. The latest Apache release is 0.2.0 (incubating); the npm package maka-agent ships real builds under the nightly tag while latest is still a placeholder, and the CLI README warns that commands and local data formats may change. macOS is the main platform; Windows and Linux are previews. API keys are stored in a plaintext credential-vault.json readable by your OS account, and an upgrade has already dropped older JSONL transcripts. Vs opencode or Claude Code: you are picking it for the audit log and replay, not for a bigger ecosystem. Stars have climbed fast for a five-month-old repo (about 290 a week since it opened in May 2026), so don't read the count as proof of adoption.

## Verdict

The agent harness to try when you need to see and replay what the agent actually did — treat it as nightly software for now.