Skip to content

Roadmap

This page tracks the near- and longer-term direction. The REST API, MCP server, and hybrid search remain first-class — everything below augments those surfaces.

Done

These have shipped and are in main.

  • Virtual filesystem export — mirrored tree with metadata.json and chunk files so traditional tools can inspect semantic data.
  • stratafs fs export CLI — exports a source's metadata view into any directory.
  • FS bridge package (pkg/fsbridge) — reusable code shared between fs export and the FUSE/WinFsp mount.
  • Wails desktop UI — control panel for managing sources, watching the queue, triggering exports. Packaged as AppImage / .deb (Linux), .pkg (macOS), NSIS installer (Windows). See Desktop App.
  • Native mounts — stratafs fs mount exposes a read-only filesystem view via FUSE on Linux/macOS and WinFsp on Windows.
  • OS search integration — GNOME Shell SearchProvider2, macOS Spotlight importer, Windows Search IFilter.
  • Contextual actions — Nautilus extension, macOS Finder Sync, Windows Explorer shell extension.
  • Enterprise connectors — SharePoint/OneDrive via Microsoft Graph, Google Drive with Workspace export, Jira issues and attachments. Incremental sync via delta APIs.
  • Desktop packaging — Linux (AppImage + .deb + systemd user service), macOS (signed .app + DMG + LaunchAgent), Windows (NSIS installer + tray + Windows Service).
  • Release automation — scripts that bundle the ONNX Runtime, binaries, and desktop UI into platform archives. See scripts/release-bundle.sh and .github/workflows/release.yml.

In flight

  • Enterprise security — RBAC for authentication and source-level permissions. Tracking issue: #TBD.

Planned

  • Streaming search results — Transfer-Encoding: chunked on /search for very large result sets.
  • Custom embedding models — first-class support for loading any ONNX-compatible model from disk, not just the BGE family.
  • Cross-source ranking signals — per-source weight, recency boost, "trusted source" pinning.
  • Quality benchmarks in CI — every PR runs research/benchmarks/ and posts a delta against main.
  • Encrypted source databases — SQLCipher-backed source DBs for at-rest encryption.

Stretch / under discussion

  • Distributed read replicas — read-only mirrors of a source's DB for higher search QPS. SQLite WAL streaming is the candidate mechanism.
  • Live retrievers — keep an open connection to long-running agents and stream new chunks as they're indexed.
  • OCR pipeline — image and scanned-PDF support, behind an opt-in flag for the additional model weight.
  • Multimodal embeddings — image + text in one vector space for searches like "find me the architecture diagram in the docs".

How we prioritize

Roughly: agent-facing > local-first > server / cloud. Anything that makes StrataFS a better backbone for a long-running coding or research assistant tends to jump the queue.

To propose something: open a GitHub issue with a one-paragraph problem statement and a concrete-enough scope that we can argue about it. Strong opinions weakly held.