apache/maka
Apache Maka (Incubating) is a high-performance agent workspace that keeps a complete record of everything it did.
What it solves
Apache Maka provides a high-performance workspace for AI agents to execute tasks. It addresses the need for transparency and reliability in agentic workflows by maintaining a complete, append-only record of every action taken by the agent, ensuring that runs are measurable and reproducible rather than based on anecdotal claims.
How it works
Maka uses a "log-as-runtime" architecture where every model message, tool call, permission decision, and termination is recorded as a RuntimeEvent. The system consists of a Runtime Host that manages the execution authority, which can be accessed via various thin clients including a Desktop app, a Terminal User Interface (TUI), and a Command Line Interface (CLI). Users provide their own models via cloud APIs, local models, or compatible gateways, and all session data and records remain local to the machine.
Who it’s for
It is designed for developers and researchers who need a controlled environment to run AI agents, benchmark their performance using official verifiers, and maintain a durable audit trail of agent activities.
Highlights
- Append-only Event Log: The entire runtime state is a projection of a durable log, enabling crash recovery and precise auditing.
- Local-First Privacy: Sessions, settings, and run records are stored locally on the user's machine.
- Model Agnostic: Supports cloud APIs, local models, and compatible gateways.
- Multiple Interfaces: Offers a Desktop application, TUI, and CLI for different interaction styles.
- Integrated Evaluation: Includes an evaluation framework to measure task completion rates and costs.
Related
- Project
- Project
- Project
- Project
- Project