using-brainpalace

# BrainPalace Expert Skill Search indexed docs / code with five modes: BM25 (keyword), Vector (semantic), Hybrid (fusion), Graph (knowledge graph), Multi (comprehensive fusion). This is the **search decision contract**. Operational reference (indexing, injection, jobs, cache, server lifecycle, per-mode deep dives) lives in the **Reference Documentation** table at the end — look it up on demand; the CLI's own `--help` is authoritative for flags. **BrainPalace is indexed for this project — search via `brainpalace query`, not Glob/Grep.** Route by query type: exact symbol/token/path → `--mode bm25` (keyword index, **no embedding round-trip** — ms latency, as fast as grep); concept → `--mode vector`/`hybrid`; relationships → `--mode graph`. Server down? Start it: `brainpalace start`; if it answers `503` "rehome pending" the project MOVED — run `brainpalace rehome --resume` (or restart; it auto-resumes) first. `brainpalace query --json` keys are `text`/`source`/`score`/`chunk_id`/`start_line`/`end_line` (no `file_path`; lines null off-code); on failure stdout is `{"error": ...}` with no `results` key and a non-zero exit — check it, never append `2>/dev/null`. ## Mode Decision Table — Quick Reference Pick `--mode` for any inline search. Default to `hybrid` when unsure. | Query type | Mode | |---|---| | General / unknown | `--mode hybrid` | | Conceptual ("how does X work") | `--mode vector` | | Exact symbol / error / token / file path | `--mode bm25` | | Relationships ("what c...