Project memory for coding agents. Single binary, hybrid search, find past decisions with citations in seconds.
cargo install mempalBM25 keyword matching + vector semantic search, merged via Reciprocal Rank Fusion. Exact-match error codes and understand semantic similarity.
Subject-predicate-object triples with temporal validity. Timeline view, stats, and relationship invalidation when facts expire.
Automatic discovery when the same room appears in multiple wings (projects). Search results include inline tunnel hints.
MEMORY_PROTOCOL (9 rules) embedded in MCP ServerInfo. Any agent learns the workflow on connect — zero configuration.
model2vec-rs (BGE-M3 distilled) as default embedder. Zero native dependencies, supports Chinese, Japanese, Korean, English.
Everything in ~/.mempal/palace.db (SQLite + sqlite-vec). Backup is cp. Migration is scp. No server process.
status, search, ingest, delete, taxonomy, kg, tunnels. Each tool carries field-level documentation in its JSON schema.
Cross-session behavioral learning. Agents record OBSERVATION / LESSON / PATTERN entries. Future sessions search and learn.
Soft-delete + purge two-step deletion. Audit logging. Semantic dedup detection. Dry-run preview mode.
query → BM25 (FTS5) → keyword ranking
→ Vector (sqlite-vec) → semantic similarity ranking
→ RRF Fusion (k=60) → merged ranking
→ Wing/Room filter → scoped results
→ Tunnel hints → cross-project references
→ Citations (drawer_id + source_file)