Highlights
Opt-in local semantic search and background daemon
ctx 0.25.0 adds an opt-in background maintenance daemon and hybrid semantic search. The daemon can keep lexical imports fresh, serve queries without reopening the index for every command, acquire the pinned intfloat/multilingual-e5-small model, and build the semantic index in bounded background work.
The preview remains disabled by default. Enable daemon-owned lexical maintenance with:
[daemon]
enabled = trueEnable both the daemon and semantic search with:
[daemon]
enabled = true
[search]
semantic = trueRun ctx setup after opting in. ctx status reports daemon state and indexing coverage. Apple Silicon can use Core ML acceleration; the release also ships managed ONNX Runtime sidecars for CPU execution across the supported platform matrix.
Faster, more useful lexical search
- Search terms now use OR semantics by default.
- Conversational user and assistant messages rank ahead of lower-value event records.
- Native refresh skips unchanged histories and avoids rescanning FTS rows.
- Bulk imports and WAL recovery are bounded more carefully on large indexes.
Broader provider and release support
- Added MiMo Code integration installers.
- Updated Factory Droid, AstrBot, and other native importers for current on-disk formats.
- Added coarse hardware capability telemetry for release planning without collecting raw hardware identifiers.
- Hardened cross-platform release construction, platform compatibility checks, and macOS Developer ID signing/notarization.
Notes
- Daemon maintenance and semantic search are preview features and remain opt-in in 0.25.0.
- Windows binaries are authenticated by signed release metadata and SHA-256 checksums but are not yet Authenticode-signed.
- Local semantic indexing downloads the pinned embedding model on first use and can take time for large histories; lexical search remains available while backfill runs.
Contributors
Thanks to @atharva-patill, @j2h4u, and @rNoz for contributions included in this release.
