AI features

Auto-tagging, semantic search, and similar-asset suggestions — opt-in per organization, human-reviewed, never used to train models.

Enabling AI

AI is off by default. An owner or admin enables it under Settings → AI. Disabling it stops all processing immediately; nothing breaks — search simply falls back to full-text.

Auto-tagging

On import, a vision model reads the asset (image + metadata) and proposes tags, a type, and a description. Suggestions go through human review — they land in the review queue at /{org}/review, and nothing changes in your library until someone approves it. Your corrections feed back into future suggestions.

Semantic search

Every asset gets an embedding (a numeric fingerprint of what it is). Searching combines two signals: 0.4 × full-text rank (bilingual EN/FR) + 0.6 × cosine similarity. That's how “dark hero with big type” finds final_v3_FINAL.png. Also available to integrations via POST /api/v1/search.

Find similar

Each asset page shows its nearest neighbors by embedding distance, with a similarity percentage — useful for finding past work on the same theme, or spotting near-duplicates before they pile up.

Privacy & data handling

• Content is processed by OpenAI only if your org enabled AI — see our subprocessor list.
• No training: your content is never used to train third-party models.
• No solely-automated decision affects you or your data — every AI suggestion is human-approved.
• Embeddings live in your org's row-isolated tables, deleted with your data (Privacy Policy).

Plans

AI features are included in the Business plan. See Pricing & limits.