Your assets. Your provenance. One platform.
Asset management on the haystak engine: provenance-aware, schemaless tagging, content-first search, and rich-media preview. Manage assets where they already live, with your own ingest, your own taxonomy, and no migration. haystak Assets is in early access: the features below are running today, and we're hardening them with beta users.
Everything you need to manage assets, none of the workflow you don't.
haystak Assets is in beta. These capabilities are running today, inherited from the haystak engine, and we're refining them with early-access users.

Unlimited tagging
No field limits, no schema waits. Tag anything, any type, any time; new metadata is instant, never a migration.
Content is first-class
Search inside the file and tag the content. Extracted text is a real entity with its own attributes.
Similarity
SimHash near-duplicate detection across the whole library, with configurable thresholds.
Auto-extraction
EXIF, XMP, PDF text, Office, OCR, media text, and thumbnails for RAW/HEIC/PSD. Deterministic and rerunnable. Media text requires the Media tier.
Renditions
One master, many renders: thumbnails and previews at ingest, content-addressed and linked by lineage.
Provenance & C2PA
Every tag carries its origin; Content Credentials are validated at ingest. We validate, never sign.
Play it. Scrub it. Search inside it.
Poster frames and scrub sprites are generated at ingest so video is previewable the moment it's indexed. haystak reads the caption text already in the file when there is one, and falls back to on-prem speech-to-text when there is not, so a search hit resolves to a start and end second in the media. Both paths run on your hardware and both require the Media tier.
- Poster frames and scrub sprites, generated at ingest and reused by content hash.
- Media text: caption tracks harvested with per-cue timecodes, or on-prem speech-to-text when there is no usable track. Speech-to-text verified on wav, mp3, aac, flac, ogg, opus, m4a. Media tier.
- Where the text came from. Every media text body records its origin: the file's own caption track, speech-to-text, or an operator override. Files are findable by that origin, so a human-authored caption and a machine transcript are never confused for each other.
Rights start with access and provenance.
The controls that protect what you manage are the same ones that keep search locked to who is allowed to see it.

RBAC-trimmed access
Search and every MCP call are trimmed to role and group grants. The same gate for people and AI.
Tamper-evident audit
Cryptographically chained record of who did what, when. The backbone of attributable asset audit, with alteration detectable, not just logged.
Read-only originals
Masters are never modified; renditions and tags live separately. Original metadata is preserved intact.
Powered by haystak.
Cross-source indexing, MCP-native, programmable ML, backend-agnostic. One install, many asset libraries.
Point haystak Assets at your library.
A walkthrough on your real assets: tagging, similarity, and rich-media preview, where they already live.