Apache Maka is a local-first AI agent workspace that records model messages, tool calls, tool results, permission decisions, and termination events as an append-only log.
The tool addresses the need for transparent, auditable AI agent execution on your own machine. Rather than sending work to remote services, Maka runs agents locally with tool execution sandboxed at a defined boundary. All interactions—what the model said, what tools it called, what those tools returned, and permission decisions made—are captured as recoverable execution facts in an append-only log. This event-sourced approach means you can inspect the complete history of an agent's work and understand exactly what happened at each step.
Maka suits developers and teams who want to run AI agents for real work while maintaining full visibility and control over execution. It is designed for those who need to inspect projects, run tools safely, and keep all data local rather than relying on external services. The project provides both a desktop application and a CLI, giving flexibility in how you interact with your agents. If you are evaluating agent runtimes and need local-first execution with transparent logging of all model and tool interactions, this tool's event-sourced architecture distinguishes it from cloud-based alternatives.
The project is under active development with the macOS Apple Silicon desktop build released as an early public version. Data formats, CLI commands, and experimental capabilities remain subject to change. The tool is incubating at the Apache Software Foundation and has not yet received full project endorsement, though this reflects its stage rather than code completeness. Windows support is available as a preview build, while Linux support is planned for future release.
