Slopweaver

A local-first, zero-key org world model — the own-your-data alternative to hosted enterprise search.

Point Slopweaver at a GitHub repo and it builds a queryable world model of your team's work, entirely on your machine: it ingests the history, runs a tiered bronze → silver → gold synthesis, and lets you ask grounded, cited questions of it. No API keys, no data leaving your machine, no accounts.

/slopweaver:onboard your-org/your-repo
/slopweaver:ask "what has the team shipped recently, and what's blocked?"

Why

• Zero-key. Nothing to configure. Language-model calls run on your existing Claude Code session (via the claude CLI — no API key, no SDK). Embeddings run on-device. The only optional credential is a GitHub token for private repos, and it defaults to your gh CLI login.
• Local-first. Your world model lives under $SLOPWEAVER_HOME on your machine. The only things that leave it are the GitHub reads you authorise and the Claude calls you already make. See docs/security.md.
• Runs anywhere. No native build step — inference is pure WASM. The same plugin works on macOS, Linux, and Windows, on any CPU architecture.
• Grounded, not hallucinated. Every answer cites the exact records it's built from; a citation the model can't back is stripped, not shown.

Install (as a Claude Code plugin)

/plugin marketplace add /path/to/slopweaver
/plugin install slopweaver@slopweaver-dev

On first session the plugin installs its own dependencies into its plugin-data dir (one-time, ~a minute) — you don't run npm install. Requires Node + Corepack/Yarn and the claude CLI (you already have it). Then run /slopweaver:onboard to build your first world model.

The pipeline

Command	Stage	What it does
/slopweaver:refresh	bronze	Ingest recent GitHub activity (PRs, issues, reviews, comments) into a local corpus. Incremental — resumes from a watermark.
/slopweaver:derive	silver	Free, deterministic synthesis: a people/container directory, a cross-ref graph, and opportunity detection.
/slopweaver:distil	gold	LLM map-reduce into grounded markdown digests (via your Claude session; caches per batch, so re-runs are cheap).
/slopweaver:ask	query	A grounded, cited answer over hybrid retrieval (on-device semantic ⊕ keyword, fail-soft to keyword).
/slopweaver:facts	query	The raw ranked record slice, no LLM — for feeding a subagent or eyeballing the corpus.

To update: re-run refresh → derive → distil. refresh pulls only new activity; distil only re-synthesises batches whose content changed.

Trust

The biggest objection to any tool that ingests company data is "where does it go?" — here, nowhere you don't already send it. A shipped hygiene gate scans every tracked file for leak classes (secrets, absolute paths, raw workspace IDs) and runs in CI + a local pre-push hook, so neither you nor a fork can commit your org's secrets. Full threat model + data-flow: docs/security.md.

Development

Requires Node (see .nvmrc) and Yarn 4 (via Corepack — corepack enable once).

yarn install
yarn slopweaver doctor  # smoke-test the CLI
yarn typecheck          # tsc --noEmit
yarn test:unit          # vitest
yarn hygiene            # public leak-class gate

Architecture: docs/architecture.md.

License

MIT — see LICENSE.