apache / maka

An event-sourced AI agent workspace where every action is an immutable log entry, making the entire runtime auditable, replayable, and recoverable.

活跃维护 Apache 2.0 TypeScript Tracked
5.9k 566 35 分钟前
CINPM
agent-runtime ai ai-agent apache cli desktop electron event-sourcing llm local-first maka tool-use typescript incubator
GitHub 对比

星标趋势

数据积累中，暂无足够数据生成趋势图

AI 分析

项目摘要

Apache Maka is an Apache Incubating project providing a high-performance, local-first AI agent workspace built with TypeScript and Electron. Its defining architectural choice is an event-sourced runtime where every model message, tool call, permission decision, and termination is recorded as an append-only RuntimeEvent, making the UI, prompt reconstruction, and crash recovery all projections of that log. It emphasizes benchmark-driven evaluation, publishing per-task results against other harnesses.

为什么值得关注

Maka's event-sourced agent runtime ('the log is the runtime') is a genuinely differentiated design choice — it enables full auditability, replayability, and crash recovery in a way that most agent frameworks don't offer. Combined with a local-first, bring-your-own-model philosophy and Apache governance, it targets a growing need for transparent, reproducible agent execution.

优势

  • Event-sourced architecture provides full audit trail, replayability, and crash recovery from an immutable log of RuntimeEvents
  • Local-first design with flexible model support (cloud APIs, local models, compatible gateways) keeps user data on-device
  • Benchmark-driven approach publishes per-task evaluation results rather than making unsubstantiated performance claims
  • Active development with 61 contributors, 14 releases in 5 months, and Apache Foundation governance providing long-term viability

局限性

  • No automated tests despite having CI — a significant gap for a project claiming high performance and production readiness
  • Still in Apache Incubator with Windows and Linux platforms listed as preview-only, limiting reach
  • README and available documentation are sparse on setup instructions, architecture details, and code examples

使用场景

  • Developers building and evaluating AI agent workflows who need full reproducibility and audit trails
  • Teams benchmarking agent harnesses and wanting transparent per-task performance records
  • Privacy-conscious users wanting local-first agent execution with the flexibility to switch between cloud and local models
目标用户： AI engineers, agent framework developers, and researchers who value reproducibility and auditability in agent execution, as well as desktop users wanting a local-first AI agent workspace.
学习曲线： 中
分析模型：LongCat-2.0 | 分析时间：10 天前