Search source · Library cutover
Loading source controls…
Iconik recovery resumes updates for mirrored assets. Native Library assets, saved selections, and Library values are retained. Library-only metadata and media coverage may differ. Pause retains the current index and keeps Iconik excluded from Library clients.
Heads up:
settings on this page are global — every change applies to every user of
the platform, immediately. There are no per-user overrides. Be careful with model picks,
weights, and job triggers.
…
Staging main
—
Production production, approved PR
—
Unreleased
—
Active assets
—
Active scenes
—
Total duration
—
Ads / Footage
—
intrinsic asset_kind
Clusters
—
Last Sync
—
Last Cluster
—
Deployed
—
Backlog · target 0
Text pending
—
done video/audio, no Phase-3 text
Embed pending
—
status = pending
Un-embedded
—
non-terminal, no embedded scene
Per-Client Coverage
| Client | Assets | Ads | Footage | Pending | Scenes | Embedded | Clustered | Transcript | OCR | Objects | Descriptions |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Loading… | |||||||||||
Daily Activity (Last 14 Days)
| Day | Added | Deactivated | New scenes | Reindex | Visualization (added vs deactivated) |
|---|---|---|---|---|---|
| Loading… | |||||
Vector Index Maintenance
| Started | Status | Indexes rebuilt | Duration | Per-index decisions |
|---|---|---|---|---|
| Loading… | ||||
Agentic Search Model & Limits
Loading…
| Environment | Model | List $ / 1M tokens (in · out) | Soft $ / search | Hard $ / search | Time limit (s) |
|---|
Per environment: the model every agent role runs on, and the spend and time limits that apply while that model is selected (blank = the environment's plain budget keys). Saves are in effect for the next run at once — nothing is cached; a run already in flight keeps what it started with. Muse needs the META_API_KEY secret on the service.
AI Refinement Model
Drift filter (cosine similarity to seed query) caps semantic quality variance across models. Latency varies — faster = chips appear sooner. OpenAI / Anthropic require their API keys to be added to Cloud Run secrets first.
Transcript Text Boost (Hybrid Ranking)
Mixes BM25-like text scores from
assets.transcript into the vector ranking.
0 = pure vector (current default before this feature).
0.5 = balanced (default).
1.0+ = text dominates.
Postgres ts_rank_cd over a GIN-indexed tsvector approximates BM25 — ~5–10ms per query.
Search V2 Engine
Controls which internal engine
/v2/search calls. Either way the response is
asset-level with matched scenes annotated. Flat is the default; switch to
hierarchical when you want cluster-aware retrieval for the V2 UI mode.
Transcripts (Gemini Flash)
Loading…
Running Now
No jobs running.
Sync Jobs
Interval (min):
| Client | Last Synced | Status | Checked | New | Copied | Inactive | Errors | Finished | Run |
|---|---|---|---|---|---|---|---|---|---|
| Loading… | |||||||||
Clustering Jobs
| Client | Status | Scenes | Clusters | Super | Started | Finished | Duration | Run |
|---|---|---|---|---|---|---|---|---|
| Loading… | ||||||||
Mixpeek Configuration
Selection persists across pages
Clustering Configuration
Loading…
Sync Errors
| Client | Asset ID | Error | Time |
|---|---|---|---|
| Loading… | |||