Problem
Today, finding components in Tangle is guesswork. Search is frontend-only and matches titles exactly, so users miss existing components and create duplicates. There's no lifecycle (active/deprecated/superseded), no team ownership, no way to curate or share libraries, and no detail view for context before adding a component.
This epic introduces a component marketplace with better search and discovery, structured publishing with lifecycle, and team-scoped libraries with sharing.
Updated direction
The current implementation has shifted from a backend-first semantic search path to an open-source-compatible frontend search baseline with optional AI enhancements.
The core reason is portability: tangle-ui should continue to work for self-hosted and open-source users without requiring a hosted backend-specific experimental API. Backend-powered semantic search can still be added as an optional hosted capability, but the default Components V2 experience should function out of the box using component sources already available to the frontend.
Current approach:
- Build a frontend component search index over available component sources.
- Support multi-source search and filtering in Components V2.
- Keep deterministic lexical/local search as the baseline retrieval layer.
- Add optional AI reranking over the already-retrieved candidate set.
- Add optional AI-generated component descriptions from known component metadata.
- Let users configure an OpenAI-compatible provider through BYOK settings: API base URL, API key, and model id.
This keeps the model out of the critical retrieval path. The LLM does not discover components from scratch; it improves ranking or generates descriptions from known candidates. That makes the feature cheaper, easier to debug, and usable without a mandatory hosted backend.
Current implementation stack
Sub-tickets
Phase 0 — Telemetry baseline
Phase 1 — Search & Discovery
OSS-compatible frontend baseline
Optional AI enhancement layer
Optional hosted/backend search path
Phase 2 — Publishing & Lifecycle
Phase 3 — Libraries, Pinning & Sharing
Problem
Today, finding components in Tangle is guesswork. Search is frontend-only and matches titles exactly, so users miss existing components and create duplicates. There's no lifecycle (active/deprecated/superseded), no team ownership, no way to curate or share libraries, and no detail view for context before adding a component.
This epic introduces a component marketplace with better search and discovery, structured publishing with lifecycle, and team-scoped libraries with sharing.
Updated direction
The current implementation has shifted from a backend-first semantic search path to an open-source-compatible frontend search baseline with optional AI enhancements.
The core reason is portability:
tangle-uishould continue to work for self-hosted and open-source users without requiring a hosted backend-specific experimental API. Backend-powered semantic search can still be added as an optional hosted capability, but the default Components V2 experience should function out of the box using component sources already available to the frontend.Current approach:
This keeps the model out of the critical retrieval path. The LLM does not discover components from scratch; it improves ranking or generates descriptions from known candidates. That makes the feature cheaper, easier to debug, and usable without a mandatory hosted backend.
Current implementation stack
Sub-tickets
Phase 0 — Telemetry baseline
Phase 1 — Search & Discovery
OSS-compatible frontend baseline
Optional AI enhancement layer
Optional hosted/backend search path
search_components/get_componentPhase 2 — Publishing & Lifecycle
Component/PublishedComponentdata modelPhase 3 — Libraries, Pinning & Sharing