elephant.md

Elephant is your searchable markdown brain — long-term memory for an AI agent, organized into documents and collections across many topics. SCOPED BY DEFAULT. With an auth token, every search defaults to YOUR library: documents and collections you own or have bookmarked (bookmarking a collection puts its documents — present and future — in your scope). Pass scope:"public" to search the whole public corpus instead. Without a token you search public documents only. The server NEVER silently widens to public — if your library has nothing relevant it says so and suggests scope:"public", so you stay in control. START HERE → ROUTE → READ. 1. Call `overview` first (pass your token) for a cheap table of contents of your collections — names, summaries, sizes, freshness. A few hundred tokens. 2. Then call `get_context(query, tokenBudget)` — it searches your scope, returns ranked summary cards, AND inlines the top documents' full content up to your token budget, so you get usable context in ONE round trip. It also returns the most relevant collections to drill into and `candidates` (cards it didn't inline) you can fetch later. 3. Responses are summaries-first: cards (tens of tokens each) by default; full content only on explicit fetch, always within a maxTokens budget. NAVIGATION TOOLS: - `overview` — your starting map (your + bookmarked collections grouped by scope, plus totals). - `get_context` — the one-call router: route-then-read within a token budget. Use this for "bring me context about X". - `find_collections` — rank your COLLECTIONS by a query, then drill into the right one. - `search_collection` — full-text + semantic search WITHIN one collection. - `search` / `discover` — document search across your scope (or scope:"public"); `discover` is natural-language/semantic. - `fetch` — one document's full markdown; pass `section` for a single heading or `maxTokens` to cap size. - `fetch_collection` — a whole collection token-efficiently (summaries, or full bodies within maxTokens). - `bookmark` — save a document, or a collection (= subscribe: its documents join your scope automatically). Prefer `get_context` for open-ended context gathering; use the discrete tools when you already know which collection or document you want. Authenticated tools take a `token` parameter (your session token from the elephant.md settings page).