Roadmap
What ships in v0.1 and what is planned next
Honest status for an early project. Use this page to set expectations, not as a commitment calendar.
Ships today
- Tus upload → S3, collections, admin UI
- Three independent modalities: multimodal chunks, speech transcripts, still frames
- Hybrid: shared segment grid → video / speech / vision embeds → weighted RRF search (query-time weights)
- Per-path process / search / costs
- Tunable multimodal
chunkDurationSec/ hybridsegmentDurationSec(5 / 10 / 15 / 30) and visionframeIntervalSec(2 / 5 / 10) - Phase 2
search-api+search-client(includingsearchHybrid) - Self-contained and external Compose deploys
- Prep/API queue split, retries, pool sizing, one-active-pipeline guards
Next (planned, not built)
| Theme | Intent |
|---|---|
| Pipeline presets | JSON config on a collection: which steps run, without a DAG UI |
| Hybrid fail-resume | Partial retry parity with speech (today UI retry = full re-segment) |
| Optional search API key | Single shared Bearer for defense in depth on busy private nets |
| Thin multi-provider | Swap embed / ASR backends per pipeline without rewriting jobs |
Explicitly deferred
- App-level admin login / public internet expose
- Visual pipeline / node editor
- Scene-cut, OCR, object detection
- Word-level timestamps / speaker diarization
- LLM answer generation / RAG chat over hits
- Global gateway RPM semaphore and multi-tenant search rate limits
If you are evaluating the project: interesting demos today are upload → pick a modality (or hybrid) → search → compare costs, and wiring a sibling backend to search-api on an internal URL.