Built for files
PDFs, docs, sheets, decks, code, text, and images return structured context.
The local database for AI agents
Three lines give your agent a private MCP database: read any file, search every workspace, and recall long-term memory without embeddings or cloud infrastructure.
Search result
Evidence pulled from files, code, and memory.
FTS5 matched the exact planning phrase and returned page-level context before the agent spent tokens opening every document.
finance/q4-plan.pdfPage 3 · score 12.8
notes/board-sync.docxParagraph 18 · updated 4 min ago
memory:user-preferencespinned · semantic
src/parser.pylines 50-80 · grep match
SQLite FTS5
Postgres tsvector
MCP native
CJK tokenization
Local-first
Zero embeddings
A new species of local agent database. Built for agents that need searchable files, durable memories, provenance, and fast workspace switching without a platform tax.
Powered by memory
Semantic, episodic, and procedural memories are ranked by relevance and recency.
Designed for speed
Deterministic full-text indexes keep retrieval fast enough to be an agent reflex.
Make every workspace searchable
Agents can search across code and documents with one tool call, then read the smallest useful page, row, or line range.
1.0 Search ->$ opendb_search "customer renewal risk"
renewal risk mentioned in Q3 actions
ARR, owner, next step, confidence
customer asks for migration timeline
Give agents memory that ages correctly
Pinned context surfaces immediately. Old facts decay. Knowledge updates supersede stale preferences while dated events stay intact.
2.0 Memory ->prefers concise answerspinned
switched backend to SQLitesemantic
old backend preferencesuperseded
Understand retrieval at scale
OpenDB favors deterministic indexing, timestamps, and provenance. Benchmark output stays explainable because every answer points back to local evidence.
3.0 Benchmarks ->LongMemEval_S R@5
token savings vs command parsing
memory stress suite
embedding calls for retrieval
“The moment an agent can read, search, and remember locally, the workspace starts to feel native.”
OpenDBLocal agent database
“Zero embeddings. Millisecond recall. The simple path finally feels like the fast path.”
Benchmark noteSQLite FTS5 retrieval
OpenDB gives agents durable local context across files, projects, and sessions.
View repository ->