waku-agent
waku-agent by Sean Chen (seanchen.io) is an open-source, local-first personal assistant framework (MIT license) designed to de-mystify modern agent architectures. Rejecting bloated, black-box frameworks that obscure how agents actually operate, Waku distills production agent engineering down to code you can read and understand in an afternoon.
The architecture clearly isolates the four foundational pillars of serious AI agents: Harness (pluggable tool execution), Loop (a crystal-clear, ~95-line Python core reasoning loop), Memory (a single, transparent SQLite database blending semantic, episodic, and procedural memory with admission gating), and Eval/LLMOps (deterministic unit tests alongside LLM-as-a-judge release gates). It pairs this with a local web dashboard that visually illuminates messages as they flow through the harness, a Telegram messaging gateway, and a custom wake-word ('Waku Waku') voice interface.
For developers and engineers who want full ownership over their personal assistant and memory without handing their data to cloud SaaS providers, Waku is an exceptional educational and practical blueprint. However, because it avoids sprawling external abstraction libraries, developers must write bespoke Python functions for custom integrations, and managing local SQLite backups remains a user responsibility.
Features
Why choose it
Trade-offs
Maintenance check
- Last commit
- 2026-11-11
- Commits, last 90 days
- 100+
- Releases, last 90 days
- 2
- Latest release
- v0.1.9 (2026-10-08)
- Open issues
- 70
- Recent issues with at least one comment
- 2/10
Oldest week on the left; the latest four weeks are orange. Source: GitHub commit activity.
Status rule: archived on GitHub = Archived; last commit within 90 days = Active; 91 to 365 days = Slowing; older = Stale. All figures come from the GitHub API on the date shown.
Most discussed open issues
- #137 The model picker offers OpenAI models that 404 on use6 comments · 2026-08-17
- #109 记忆的总结是不是过于频繁,以及不够准确。3 comments · 2026-08-13
- #338 Gemini: streamed tool calls drop thought_signature, dashboard turns 4002 comments · 2026-10-30
- #337 DeepSeek: reasoning_content is never sent back, tool turns and follow-ups 4002 comments · 2026-10-30
- #101 Add deterministic WhatsApp webhook security tests2 comments · 2026-08-11
Issue titles are quoted from GitHub. Open the link to check whether a problem still applies to the version you use.
Compatibility
Quick start
Use cases
Alternatives
Comments
No comments yet. Be the first!