Hybrid search that just works
One SQL query fuses FTS5 BM25, vector cosine similarity, and metadata scoring. Toggle the weights, switch to pure FTS or vector — same fast SQLite engine.
Learn more →
v0.2.1 · MIT licensed · 35+ file types
Hybrid RAG over your files — local, offline, MCP-native.
One query fuses BM25, vector, and metadata ranking across Local, S3, GCS, Azure, SharePoint, Drive and Jira — no cloud, no API calls. A native Model Context Protocol server hands the results to any AI agent or assistant. Open source, MIT licensed, runs on your laptop. (Still a semantic filesystem underneath.)
$ strata search "how do we rotate the signing keys" --top 3
# fused: FTS5 BM25 · vector cosine · metadata recency
1. ops/runbooks/key-rotation.md score 0.94
…rotate the Ed25519 signing key every 90 days via `strata…
2. docs/security/secrets.mdx score 0.81
…KMS-backed rotation; the signing material never leaves…
3. infra/terraform/kms.tf score 0.76
…resource "aws_kms_key" "signing" enable_rotation = true…
3 hits · 12,481 files · 41 ms · offline · MCP :8081 ready- 35+file types parsed
- 7storage backends
- 50–100 files/secindexing throughput
- < 100 mssearch latency (10k files)
Live demo
One query. Three ranking modes. Same SQLite engine.
Switch between Hybrid, FTS, and Vector to see how StrataFS fuses BM25, cosine similarity, and metadata into a single ranked list.
Try:
- auth/middleware/refresh.golocal · code0.86
// RefreshToken validates the refresh JWT and issues a new access token. func RefreshToken(ctx context.Context, raw string) (*Tokens, error) {Matches "JWT refresh" semantically + filename contains "refresh". - docs/runbooks/auth-rotation.mds3://acme-docs0.78
When the access token expires we exchange the long-lived refresh token at /v2/auth/refresh — see the sequence diagram below.FTS hit on "refresh token" + semantic match on the rotation flow. - internal/session/jwt.golocal · code0.69
// JWT signing keys are rotated daily. Refresh tokens persist 30 days.Cross-file context: JWT + refresh policy.
Canned demo — results are illustrative. Real StrataFS runs locally and indexes your own files.
What it does
A storage layer that understands what's inside your files.
Native MCP server
Model Context Protocol on port 8081, ready for Claude Desktop, ChatGPT plugins, and custom agents. No glue code, no API gateway, no SaaS bridge.
Learn more →Seven storage backends
Local + S3 + GCS + Azure Blob + SharePoint + Google Drive + Jira — one factory pattern, one query interface, separate isolated databases per source.
Learn more →Read-only by design
StrataFS never modifies your files. Mounts every source read-only and keeps all index state in a parallel .stratafs/ directory. No surprise mutations.
Embedded, not embedded-into
SQLite with FTS5 + sqlite-vec. No Elasticsearch cluster, no Pinecone account, no separate vector store. The database is a file, on your disk, that you own.
Cross-platform desktop & CLI
Native installers for macOS, Linux, Windows. Wails desktop UI. Spotlight / Nautilus / Windows Search integration. FUSE mount exposes the index as a filesystem.
How it works
A pipeline from your storage to your search bar — and your agent's context window.
Storage Local · S3 · GCS · Azure · SharePoint · Drive · Jira
Job queue fsnotify + delta-API polling · retry · recovery
Parsers 35+ types · streaming chunkers
SQLite index FTS5 BM25 + sqlite-vec + metadata
REST + MCP :8080 REST · :8081 MCP for AI agents
- 01
Connect storage
Point StrataFS at one or more sources — local directories, S3 buckets, GCS, Azure, SharePoint, Google Drive, or Jira projects. Read-only credentials only. Nothing leaves your machine.
- 02
Watch and queue
fsnotify monitors local paths in real time. Cloud sources poll via delta APIs. Changes land in a SQLite-backed job queue with retry, priority, and recovery on restart.
- 03
Parse, chunk, embed
35+ parsers extract text from PDFs, DOCX, code, spreadsheets, markup, and config files. Streaming chunkers keep memory flat. FastEmbed + ONNX runs embeddings locally — no API calls.
- 04
Search & serve
One SQL query fuses FTS5 BM25, vector cosine, and metadata into a single ranked list. Exposed via REST (port 8080) and a native MCP server (port 8081) for AI agents.
Storage backends
Seven backends. One query interface.
Read-only credentials. Per-source isolated SQLite databases. Real-time for local, polled with delta APIs for the cloud.
-
Local filesystem
fsnotify-driven, instant updates when files change.
Real-time -
Amazon S3
Polling-based sync, IAM-friendly read-only credentials.
Polled -
Google Cloud Storage
Service account JSON, bucket-level scoping.
Polled -
Azure Blob Storage
Account key or SAS, container-level scoping.
Polled -
SharePoint / OneDrive
Microsoft Graph delta API, enterprise-ready.
Polled -
Google Drive
OAuth2 + native Docs export.
Polled -
Jira
Issues, descriptions, attachments via REST API.
Polled
Get started
Install in 30 seconds. Index in minutes.
Pick your package manager. stratafs config init writes a
sensible default config; stratafs serve brings up the REST
API on port 8080 and the MCP server on port 8081.
npm install -g stratafs
stratafs config init
stratafs servemacOS · Linux · Windows
How it compares
StrataFS vs. the alternatives.
| Feature | StrataFS | Elasticsearch | Pinecone | grep / ripgrep |
|---|---|---|---|---|
| Self-hosted, MIT licensed | ✓ | ✓ | — | ✓ |
| Runs entirely offline | ✓ | ✓ | — | ✓ |
| Full-text + vector in one query | ✓ | ~ | — | — |
| Indexes Local + S3 + GCS + Azure | ✓ | ~ | — | — |
| Native MCP server for AI agents | ✓ | — | — | — |
| Zero ops: SQLite on disk | ✓ | — | — | ✓ |
| Real-time file change indexing | ✓ | ~ | — | — |
| FUSE / Spotlight / Explorer integration | ✓ | — | — | — |
Comparison reflects features as of June 2026. See individual /compare pages for nuance.
Latest writing
Field notes from semantic-search land.
StrataFS vs. Elasticsearch for semantic code search — when each makes sense
Elasticsearch is the workhorse of full-text search. StrataFS is the new workhorse of embedded semantic search. Here's an honest comparison for developer-focused use cases.
Claude Desktop + StrataFS: a five-minute MCP setup for your codebase
Wire StrataFS into Claude Desktop in five minutes. Hybrid search across your code and docs becomes a tool Claude knows how to call. Step-by-step, with the config and the troubleshooting.
Index S3, GCS, and Azure Blob buckets with semantic search
Indexing cloud storage used to mean pipelines, copies, and a separate search service. StrataFS reads buckets in place with read-only credentials and exposes a hybrid search across all of them.
Self-hosted semantic search: local AI indexing without SaaS lock-in
Most 'AI search' tools want your data on their servers. We didn't want to send our files anywhere, so we built StrataFS. Here's the case for self-hosted semantic search.
Questions
Frequently asked
What is StrataFS, in one sentence?
StrataFS is an open-source semantic filesystem that indexes Local, S3, GCS, Azure, SharePoint, Google Drive, and Jira sources, runs hybrid full-text + vector search over them, and exposes the result as a Model Context Protocol server for AI agents.
Who is StrataFS for?
Developers who want better-than-grep search over their codebase, AI engineers who need a Model Context Protocol-native retrieval layer, and enterprise teams who want semantic search across SharePoint, Drive, and Jira without a SaaS vendor in the data path.
What's the license?
MIT. The full source is on GitHub at github.com/neul-labs/stratafs. No usage limits, no telemetry, no 'we may change this license later' clause.
How mature is the project?
Version 0.2.1 as of June 2026, active development. Production-ready for single-user and small-team deployments; enterprise RBAC and at-rest encryption are in active development for the next release.
What's hybrid search and why does it matter?
Hybrid search fuses BM25 (full-text), vector cosine similarity (semantic), and metadata signals (recency, filename match, file-type) into one ranking. It catches the queries pure full-text misses (synonyms, intent) AND the queries pure vector misses (exact identifiers, error codes). StrataFS does this in a single SQL query.
Can I do FTS-only or vector-only search?
Yes. The 'mode' parameter on /search switches between 'hybrid' (default), 'fts' (BM25 only), and 'vector' (cosine only). Use 'fts' when you know the exact phrase, 'vector' when you want semantic exploration.
Explore StrataFS
Everything, one hop away.
Quickstart
Install, add a source, serve, search, and connect an AI agent over MCP — running locally in about ten minutes.
Learn more →Architecture
How the ingest pipeline, queue, parsers, embeddings, SQLite database, and REST + MCP APIs fit together.
Learn more →For developers
Codebase search that beats grep on intent queries — find code by what it does, not just the strings it contains.
Learn more →For AI engineers
Retrieval built for LLMs: hybrid ranking, chunk-aware results, and a native MCP server your agents can call.
Learn more →For enterprise
Index SharePoint, Google Drive, and Jira alongside cloud object storage — self-hosted, no SaaS lock-in.
Learn more →Compare
How StrataFS stacks up against Elasticsearch, Pinecone, and grep — and when each alternative is the better fit.
Learn more →Glossary
Plain-English definitions of hybrid search, BM25, vector search, embeddings, sqlite-vec, FTS5, chunking, MCP, and RAG.
Learn more →FAQ
Common questions about installation, search behavior, the MCP server, storage backends, and day-to-day operations.
Learn more →About
Project goals, current status, and the roadmap for StrataFS as an open-source, local-first semantic search server.
Learn more →Index your storage. Search it like a human.
MIT-licensed. Multi-storage. Native MCP server for AI agents. Local, S3, GCS, Azure, SharePoint, Drive and Jira.