Open-source AI personal assistant for planning, writing and personal knowledge, runnable on a Mac or self-hosted with your own model keys.
ShenSeanChen/waku-agent
A local-first personal AI assistant in readable Python, with its own agent loop, SQLite-based memory, and built-in evals.
Open-source alternative to:ChatGPT
- Stars
- 2,082
- Forks
- 410
- Open issues
- 57
- Last commit
- 11 days ago
What is it?
waku-agent is a personal AI assistant that runs on your own machine. It is built around four parts: a harness that connects gateways such as a terminal, a browser dashboard, Telegram, and Discord to one agent loop; a loop of about 95 lines of Python that reasons, acts, and repeats; a memory system covering semantic, episodic, and procedural (skill) memory; and an eval layer that combines deterministic tests with LLM-as-judge scoring.
Memory is stored in a single SQLite file, and a gate decides whether each message is worth remembering while a consolidation pass decides what to keep. The project supports several model providers, including Anthropic, OpenAI, Gemini, DeepSeek, and OpenRouter, selected through an environment variable and an API key. An optional integration connects the agent to the hosted Waku Memory service so the same memory can be shared with other agents.
The code is intended to be read and modified rather than used as a black box. It is MIT licensed, except for the hosted/ directory, which uses Elastic License 2.0.
Key features
- 01Agent loop of about 95 lines of plain Python with reasoning, action, and repetition
- 02Semantic, episodic, and procedural memory stored in a local SQLite file
- 03Terminal, browser dashboard, Telegram, Discord, and voice gateways
- 04Deterministic tests and LLM-as-judge evals with a release gate
- 05Support for multiple model providers including Anthropic, OpenAI, Gemini, and OpenRouter
Use cases
- Running a personal assistant on your laptop that keeps its memory locally
- Studying how an agent loop, memory retrieval, and evals fit together in a small codebase
- Forking the harness to add a new messaging gateway or tool
- Sharing one memory across Claude Code, Codex, and other agents via Waku Memory
Quick start
- 1
Install the package from PyPI
pip install waku-agent - 2
Talk to your Waku in the terminal (it will ask for an API key the first time)
waku - 3
Or open the browser dashboard at localhost:7777
waku dashboard
Things to know
- Needs an API key from a model provider (e.g. Anthropic, OpenAI) to run.
Topics
FAQ
- What is waku-agent?
- A local-first personal AI assistant in readable Python, with its own agent loop, SQLite-based memory, and built-in evals. waku-agent is a personal AI assistant that runs on your own machine. It is built around four parts: a harness that connects gateways such as a terminal, a browser dashboard, Telegram, and Discord to one agent loop; a loop of about 95 lines of Python that reasons, acts, and repeats; a memory system covering semantic, episodic, and procedural (skill) memory; and an eval layer that combines deterministic tests with LLM-as-judge scoring.
- Is waku-agent free to use? What license does it have?
- waku-agent is open source under the MIT license. Check the license terms before using it in commercial products.
- Is waku-agent still maintained?
- Actively maintained. The last commit to waku-agent was on Oct 11, 2026.
- What can I use waku-agent for?
- Running a personal assistant on your laptop that keeps its memory locally Studying how an agent loop, memory retrieval, and evals fit together in a small codebase Forking the harness to add a new messaging gateway or tool Sharing one memory across Claude Code, Codex, and other agents via Waku Memory
- How do I install waku-agent?
- Install the package from PyPI: pip install waku-agent Talk to your Waku in the terminal (it will ask for an API key the first time): waku Or open the browser dashboard at localhost:7777: waku dashboard
- Where can I get waku-agent?
- The source code and installation instructions are on GitHub: https://github.com/ShenSeanChen/waku-agent
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
A lightweight, self-hosted personal agent app with a TUI and GUI, local workspace tools, human approval, persistent sessions, memory, and Skill/Plugin/MCP extensions.
Self-hosted, local-first personal AI agent with layered memory, tool permissions, workflows and IM channel integration.
A self-hosted platform of persistent AI agents with long-term memory that collaborate and answer across chat channels, run in a single container.
A documented architecture and installable Claude Code plugin for building a persistent AI chief of staff with memory, daily briefings and scheduled background jobs.
Local-first personal OS agent built on Hermes that logs mood, sleep, meals and habits, then surfaces patterns across them over time.