Apache Maka

Overview

Apache Maka is a local-first AI agent workspace incubating at the Apache Software Foundation. Where most agent tools focus on the model interaction, Maka focuses on the execution record — every model message, tool call, tool result, permission decision, and termination event is captured in an append-only log that is never deleted. This makes agent sessions reproducible, recoverable, and auditable. The workspace provides three surfaces: an Electron+React desktop application for visual interaction, a terminal CLI/TUI for headless and scripting workflows, and an evaluation framework for systematic agent testing.

Key Capabilities

Durable execution logging: Maka’s defining feature is its append-only log. Every interaction between the model and tools is recorded with full fidelity — not just the final output, but the sequence of tool calls, their results, permission decisions made during execution, and termination events. This log serves multiple purposes: crash recovery (interrupted sessions resume from exactly where they stopped), session branching (fork a session at any point to explore alternatives), and audit trails (review exactly what an agent did and why).

Multi-surface workspace: The same workspace state is accessible through three interfaces. The Electron desktop app provides a visual, graphical interaction experience. The terminal CLI/TUI enables headless operation, scripting, and integration into existing development workflows. The evaluation framework supports systematic testing of agent behavior across defined scenarios.

Bring-your-own-model: Maka is model-agnostic. Connect it to a cloud API (Anthropic, OpenAI, Google), a local model (Ollama, llama.cpp), or any compatible gateway. The workspace handles the orchestration layer while the model choice remains yours.

Session management: Sessions in Maka are first-class objects that can be created, archived, searched by content, renamed, retried from specific points, regenerated with different parameters, and branched to explore alternative approaches. This goes beyond simple conversation history — sessions are structured execution records with full replay capability.

Use Cases

Developers who want reproducible agent workflows use Maka’s append-only log to ensure that every coding session can be reviewed, replayed, or branched. Teams evaluating different models or prompting strategies use the evaluation framework to run systematic comparisons across the same task set. Security-conscious organizations use the sandbox boundary and permission logging to maintain audit trails of what agents accessed and modified. Solo developers use the desktop app for interactive work and the CLI for scripting repetitive agent tasks.

Considerations

Apache Maka is an incubating project at the Apache Software Foundation — it has institutional backing and governance but is not yet a top-level Apache project. Primary support is macOS Apple Silicon; Windows has an unsigned preview build, and Linux is not yet supported. At 3.6K stars, it is earlier-stage than established alternatives like Claude Code or Codex, though the Apache Foundation backing provides confidence in long-term maintenance. The local-first architecture means no cloud dependency but also no built-in cross-device sync.

Who It’s For

Developers who value execution reproducibility and auditability in their agent workflows. Teams that need to evaluate agent behavior systematically across scenarios. Organizations with compliance requirements that mandate audit trails for AI-assisted development. Anyone who has lost work to a crashed agent session and wants guaranteed recovery.