Wardwell

v0.5 — semantic search

Your AI searches everything you know.

Every decision, every fix, every pattern you've discovered — indexed and searchable by your AI agent mid-session. It finds what you forgot you knew.

Install
$ brew tap yakschuss/wardwell
$ brew install wardwell
$ wardwell init
# registers MCP server, builds search index
How it works

You work. Wardwell indexes.

Decisions, history, blockers, lessons — written to a plain-text vault as you go. Full-text indexed automatically. Markdown and JSONL you can read yourself.

The AI searches without asking.

Wardwell exposes search as MCP tools. When your agent needs context — a past decision, a pattern from another project, that fix from three weeks ago — it queries the index directly.

Cross-project. Cross-session.

Nothing stays siloed. A lesson learned in one project surfaces in another. Your vault is one search space — every project, every session, every insight.

What agentic search looks like
// agent calls wardwell_search mid-session query: "KV batch write timeout" action: search // 3 results across 2 projects lessons.jsonl — Root cause: unbounded batch size Prevention: chunk KV writes to 25 keys max decisions.md — Switched to sequential writes for reliability Tradeoff: 2x slower, zero timeouts history.jsonlFeb 18 — Fixed KV timeout in vault sync
What's new in v0.5

Search by meaning, not just keywords.

Ask "how did we handle auth failures?" and get results even if those exact words never appear. Semantic search runs locally — no API calls, no data leaves your machine.

Zero setup. Just use your folders.

Drop files in work/, personal/, whatever — Wardwell figures out the domain from folder structure. No tagging, no frontmatter, no config.

Remember what you shipped.

At the end of a session, Wardwell prompts you to log a one-liner about what you built. Next time the AI picks up where you left off.

Get notified when new search capabilities ship.