v0.1.0 — the first tagged release

The first tagged release. waku has been usable for a while — this marks a point people can pin, and credits the contributors who got it here.

What waku is

A local-first personal assistant, in code you can read in an afternoon. Four pillars, none of them hidden behind a framework: Harness · Loop · Memory · Eval/LLM-Ops. Your memory is one SQLite file on your own machine.

git clone https://github.com/ShenSeanChen/waku-agent && cd waku-agent
uv venv && uv pip install -e .
cp .env.example .env          # pick a provider, paste one key
uv run waku                   # or: uv run waku dashboard

In this release

Talk to it from anywhere — CLI, a local dashboard at :7777, voice with a wake word, Telegram, and now Discord.

Memory that decides for itself — semantic (FTS5), episodic and procedural, with a retrieval gate that decides whether to remember and a consolidation pass that decides what to keep. Episodes can live in SQLite or Notion.

Bring your own model — Anthropic, OpenAI, Gemini, DeepSeek, MiniMax, Kimi, GLM, xAI or OpenRouter. One dialect in the loop, a small adapter for the rest, and you can switch model live from the dashboard.

The Arena — race one message through every model at once, in isolated homes, and score them on completion, an LLM referee, latency and real cost. Each model's knowledge-cutoff date is shown, so stale world-knowledge doesn't read as low capability.

Evals built in — deterministic tests and LLM-as-judge side by side, behind a release gate.

Calendar — local SQLite + .ics stay the source of truth, with opt-in write-through to Apple Calendar or Google Calendar.

Experimental: sub-agent delegation — behind WAKU_EXPERIMENTAL=1, waku can hand a coding job to pi running on the same model, and stream its work into the dashboard live. Its tokens land on the same ledger, so a delegated run isn't free.

Thanks

Community contributions in this release, from eight people who mostly opened their first PR here: gateways, encoding fixes across the whole tree, the trace viewer and its multi-day mode, a Notion memory backend, Google Calendar sync, model knowledge-cutoffs, and the first community skill.

@Tian-Timm · @wilenWang · @drakegriffith · @luminary19 · @c436zhan · @likweitan · @Dodothereal · @floze-the-genius · @epicnellson · @junlov · @maanas1234

Known gaps

• waku/ops/dashboard.py is a 1,695-line god file. Splitting it is the next piece of work.
• Google Calendar auth still goes through the gcloud CLI; a real in-app OAuth flow is #31.
• Integrations are configured by hand-editing .env. A proper Connections surface is #32.

MIT. Built alongside Sean's AI Stories.