Visual retrieval
Ground on-screen evidence
Slides, charts, product demos, whiteboards, and screen recordings become searchable context for downstream reasoning.
Cerul turns slides, charts, demos, code screens, and whiteboards into queryable evidence for agents. One platform backbone, two tracks: lightweight b-roll discovery and knowledge-dense retrieval.
Shared backbone
2 tracks
B-roll and knowledge retrieval on the same platform spine.
Public API
2 endpoints
Search and usage first. Heavy processing stays behind workers.
Agent path
Skill-first
Installable skills and direct HTTP before extra adapters.
Interactive Demo
Surface
Segment retrieval
Output
Summary plus time range
Active Input
Find the segment where the speaker explains the AGI timeline and shows a roadmap slide.
Response
Running initial preview request...
Try it now
curl "https://api.cerul.ai/v1/search" \
-H "Authorization: Bearer YOUR_CERUL_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"query": "Sam Altman views on AI video generation tools",
"search_type": "knowledge",
"max_results": 3,
"include_answer": true
}'{
"results": [
{
"id": "yt_hmtuvNfytjM_1223",
"score": 0.96,
"title": "Sam Altman on the Future of AI Creative Tools",
"description": "OpenAI CEO discusses the implications of AI video generation for creative industries",
"video_url": "https://www.youtube.com/watch?v=hmtuvNfytjM&t=1223s",
"thumbnail_url": "https://i.ytimg.com/vi/hmtuvNfytjM/hqdefault.jpg",
"source": "youtube",
"speaker": "Sam Altman",
"timestamp_start": 1223,
"timestamp_end": 1345,
"duration": 122,
"answer": "Altman believes AI video generation will democratize filmmaking, allowing anyone to create professional content. He emphasizes the importance of human creativity in prompting and curation."
},
{
"id": "yt_8XJ6z1K3n9P_445",
"score": 0.91,
"title": "Fireside Chat: AI and the Future of Media",
"description": "Discussion on how AI tools are reshaping video production and storytelling",
"video_url": "https://www.youtube.com/watch?v=8XJ6z1K3n9P&t=445s",
"thumbnail_url": "https://i.ytimg.com/vi/8XJ6z1K3n9P/hqdefault.jpg",
"source": "youtube",
"speaker": "Sam Altman",
"timestamp_start": 445,
"timestamp_end": 582,
"duration": 137
}
],
"credits_used": 2,
"credits_remaining": 998,
"request_id": "req_know_20240310_xyz"
}curl "https://api.cerul.ai/v1/search" \
-H "Authorization: Bearer YOUR_CERUL_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"query": "cinematic drone shot of coastal highway",
"search_type": "broll",
"max_results": 3
}'{
"results": [
{
"id": "pexels_28192743",
"score": 0.94,
"title": "Aerial drone shot of coastal highway",
"description": "Cinematic 4K drone footage of winding coastal road",
"video_url": "https://videos.pexels.com/video-files/28192743/aerial-coastal-drone.mp4",
"thumbnail_url": "https://images.pexels.com/photos/28192743/pexels-photo-28192743.jpeg",
"duration": 18,
"source": "pexels",
"license": "pexels-license"
}
],
"credits_used": 1,
"credits_remaining": 999
}Why Cerul
Agents can already search web pages. The missing layer is visual evidence inside videos. Cerul indexes what was actually on screen, not only what happened to be spoken.
Visual retrieval
Slides, charts, product demos, whiteboards, and screen recordings become searchable context for downstream reasoning.
Thin orchestration
FastAPI handles auth, usage, and response shaping. Ingestion, indexing, and media-heavy work stay in Python workers.
Replaceable models
CLIP, OpenAI embeddings, or future internal models can slot into the same shared retrieval backbone.
Two tracks, one platform
The front end borrows Tavily's unified product rhythm: one brand, one stack, distinct product surfaces.
Launch track
A lightweight showcase that proves the value of visual search quickly, with lower ingestion cost and immediate demo value.
Core moat
Long-form talks, podcasts, product keynotes, and technical videos indexed into segments that reflect what was said and shown.
Benchmarks
Cerul leans on benchmarks to establish trust, especially where transcript-only systems lose visual recall.
Slide recall
How well the system captures text and chart evidence from frames
93%
Demo grounding
Whether product screens and on-screen actions remain queryable
88%
Transcript-only gap
How much signal is lost if the system reads words but ignores visuals
71%
Operator console
Usage ledger
128 credits
Monthly credit accounting with the same request IDs used for search logs.
Index freshness
3h 14m
Current lag between source discovery and available search results.
Search health
99.94%
Healthy request success rate across both product tracks.
Pricing
Free
$0
for early evaluation
Best for trying the public API surface and validating early demo integrations.
Builder
$20
per month
For teams building agent workflows that need predictable usage and more active keys.
Enterprise
Custom
volume and support matched
For production deployments with private ingestion pipelines, SLA expectations, and compliance review.