August 22, 2026Open SourceInfrastructureAgents

Apache Maka: Agent Infrastructure Grows Up

A quiet milestone on GitHub Trending today: apache/maka, a local-first AI agent workspace, incubating at the Apache Software Foundation. Around 2,000 stars and climbing fast, but the star count isn't the story. The ASF stamp is. Agent runtimes have so far lived in personal repos and startup orgs — this is one entering the same governance machinery that raised Kafka, Spark and Airflow.

What Maka actually is: a workspace where sessions, settings and run records stay on your machine by default. Desktop app (Electron plus React), a terminal UI, a CLI, the standard tool belt of Read, Write, Edit, Bash, Glob, Grep, and granular permission controls. TypeScript, SQLite, Apache 2.0. macOS Apple Silicon is the first public build.

The design choice that matters is event sourcing. Model messages, tool calls, tool results, permission decisions and termination events are recorded as an append-only log — a recoverable execution history that exists independently of whatever survives in the model's context window. Anyone who has watched an agent compact its context and forget why it did something will recognize what this is: a flight recorder. The model's memory can be lossy because the workspace's memory isn't. You can search, retry and branch sessions from the log, and audit exactly which permission decision let which tool call happen.

The bet underneath is that agent execution history is infrastructure, not app state — too important to live inside any single vendor's cloud. That's precisely the kind of boring, load-bearing claim that foundations exist to steward. Watch whether the incubation attracts the corporate contributors that pattern usually does.

Repo: https://github.com/apache/maka
← Previous
Anthropic Ships Mythos 5 to Defenders, Behind a Filter
Next →
EnvHarness: Google Trains the World, Not the Agent
← Back to all articles

Comments

Loading...
>_