● Daemon v0.8.4 · 935 tests green

wotw writes the wiki.
Then signs it.

Watcher is a local-first AI librarian. Point it at your documents, transcripts, and code. It builds a wiki where every paragraph traces back to its source through a verifiable hash chain. Bring your own model. Nothing leaves your machine unless you say so.

npm i -g @3030-labs/wotw  ·  Node 20+  ·  AGPL-3.0
wotw · workbenchdaemon live
sources 3
research/raw/rate-limiting.md
a3c4·7e2b
transcripts/2026-05-12.txt
91de·04a8
src/limiter.ts
4f7c·19a0
wiki · compiling ~/wiki/rate-limiting.md
compiled · 3 sources · every line cited

Rate limiting

The limiter uses a token bucket sized to the burst ceiling, refilled on a fixed interval.Under sustained load it sheds requests rather than queueing them.

The window was chosen on the 2026-05-12 sync to match the upstream provider’s advertised quota, not a guess.

provenance append-only
ingest rate-limiting.md · sha256:a3c4·7e2b
extract 6 facts → facts.db
compile wrote rate-limiting.md · 3 pages · $0.030 · 35.5s
seal chain +1 · sha256:c19f·a481
ready

Confident is easy. Provable is the job.

Most AI tools ask you to trust the answer. Watcher hands you the receipt. Nothing enters the wiki without a chain entry, and nothing leaves it without citing the chain.

The pipeline

Ingest. Extract. Compile. Seal.

01 / Ingest

Watch the sources

Drop files into a folder. The daemon polls, chunks, and pages them.

+1 record / chunk
02 / Extract

Atomic claims

One fact hash per assertion lands in facts.db.

82–96% fewer tokens
03 / Compile

Assemble the wiki

Pages cluster from facts. Every paragraph cites the chain it came from.

backlinks + inline cites
04 / Seal

Sign & ship

Each operation appends a signed record. Pack any slice and hand it off.

verify --no-daemon ✓
Local-first

Runs on your machine. Answers to your audit.

The daemon runs at the edge. Your wiki is markdown on disk. Your facts are a store you can grep. Bring any model, your key is injected at spawn-time and never persisted.

AnthropicOpenAIGeminiOllama
One substrate, two doors

Memory for agents. A wiki for you.

For people

A wiki that reads like a person wrote it

Drop in PDFs, transcripts, and code. Watcher compiles a clean, backlinked wiki, and every paragraph carries an inline citation to the source it was built from.

  • Structured pages with backlinks
  • Inline citations resolve to exact source chunks
  • Read it in your browser or your editor
For agents

A memory tier that knows what it knows

Agents query atomic facts over MCP instead of stuffing whole documents into context. 96% fewer tokens per answer, and every fact returns with a hash you can verify.

  • query_facts over 17 MCP tools
  • One fact hash per atomic claim
  • Grounded answers, never a fact without a source
Pricing

Free to self-host. Hosted when you want it.

Community
$0
Self-hosted, AGPL, forever. You run the daemon, you bring the model.
Download
Pro
$29.99 /mo
Hosted, always-on daemon. Web provenance dashboard. Multi-device.
Start Pro
Compliance
$199 /mo
Retention policies, immutable audit, redaction log, signed Compliance Packs.
See Compliance

Founding rate $20/mo locked for life within the first 30 days.

Five minutes to your first cited page

Hand it the raw. Get back the record.