The lab is where Waku meets other projects: another agent, a model or a memory
product, and how Waku's agent, memory and skills connect to it. Video work
starts here. None of it is the product: nothing under Prometheus/ or evals/
imports from lab/, and lab/ never ships to PyPI.
| Topic | The question |
|---|---|
| kimi-k3/ | What makes Kimi K3 fast and cheap, and how does it do on Waku's own tasks? |
| pi-agent/ | Who owns an agent's context: the vendor, or you? |
| memory-native/ | What do mem0, Zep, LangMem and Supabase do on their own terms, before any comparison with Waku? |
| one-memory-every-agent/ | Can one memory follow you across Grok Bot, Muse, Claude Code, Codex and the Waku agent? |
- Copy _template/README.md to
lab/<topic>/README.md. Name the folder for the topic, never for a video or a date. - Fill in all six sections and keep their headings.
evals/deterministic/test_rulebook.pychecks the headings and theVerified against:line. - Keep the topic's code and any long write-up (
write-up.md) in the same folder. Boards go in as PNG screenshots, inscreenshots/; their editable.excalidrawsources and the scripts that draw them stay out of the repo.