Cerul
The video search layer for AI agents. Search video by meaning — across speech, visuals, and on-screen text.
Last verified:
What is Cerul?
Cerul is the video search layer for AI agents, enabling semantic search across video content by understanding speech, visual scenes, and on-screen text. Unlike traditional video search that only indexes transcripts, Cerul indexes what is shown—including slides, charts, product demos, code walkthroughs, and whiteboard sketches—making visual evidence searchable for AI agents.
The platform provides a single API endpoint that returns timestamped, grounded results with relevance scores, source metadata, keyframe previews, and tracking URLs that redirect to the original video. Users can optionally get AI-generated summaries grounded in matched evidence by setting include_answer=true. The API supports filtering by speaker, source, publication date, and video duration, plus an optional LLM-based reranking mode for higher relevance using the Jina Reranker.
Cerul is built for developers and teams building AI agents that need video understanding capabilities. It works with Claude Code, Cursor, Windsurf, and other MCP-compatible clients through official plugins and SDKs. The tool is fully open source, allowing teams to self-host the entire stack for free, or use the hosted API to skip infrastructure setup. New accounts receive 100 signup credits and 10 free searches per day without requiring a credit card.
Cerul pricing
Pricing model: Freemium
Free tier: $0 with 100 credits on signup plus 10 free searches every day, full public search API access, and community support. Pay-as-you-go: $8 per 1,000 credits with no monthly fee, minimum 1,000 credits per purchase, auto-recharge available, and 10 free searches daily. Pro: $29.90/month including 5,000 credits per month (effective rate $5.98/1K), top-up at $8/1K for more credits, higher rate limits, 10 free searches daily, and priority email support. Enterprise: Custom pricing for production deployments with private indexing, custom volume pricing, dedicated onboarding, enterprise security and SLAs. All plans include 100 initial free credits and 10 free searches per day. A standard search costs 1 credit; searches with include_answer=true cost 2 credits.
Cerul pros
- Searches video by meaning across speech, visuals, and on-screen text in one API call
- Indexes visual content like slides, charts, demos, and whiteboard sketches
- Returns timestamped results with relevance scores from 0.0 to 1.0
- Provides keyframe URLs for preview images of matched segments
- Includes full ASR transcript text for speech-backed segments
- Optional AI-generated summaries grounded in matched evidence
- Supports filtering by speaker, source, date, and duration
- LLM-based reranking mode available with Jina Reranker for higher relevance
- 10 free searches per day without credit card required
- 100 free signup credits for new accounts
- Fully open source with free self-hosting option
- Official TypeScript SDK and Python SDK available
- Works with Claude Code via official plugin
- MCP-compatible with Claude Desktop, Cursor, Windsurf
- 0% uptime SLA guarantee for enterprise plans
- Tracking URLs redirect to original video sources
- Speaker detection included in response metadata
Cerul cons
- Standard search limited to 400 character query maximum
- Pay-as-you-go requires minimum 1,000 credits per purchase
- Free tier has no monthly included credits
- Rate limits are standard on free and pay-as-you-go tiers
- include_answer doubles credit cost from 1 to 2 credits
- Requests blocked after daily free searches when out of credits
- No auto-recharge on free tier
- Enterprise features require custom pricing and sales contact
- Keyframe URLs only available when available (not guaranteed)
- Transcript field is null for visual-only segments
- No native mobile app mentioned
- Self-hosting requires own infrastructure setup
- Jina Reranker usage may incur additional costs
- Daily free searches reset at UTC day boundary
- No phone support listed, email only at [email protected]
Frequently asked questions about Cerul
How do credits work?
A standard search costs 1 credit. A search with include_answer=true costs 2 credits. Every user gets 10 free searches per day that do not require credits. Credits are consumed from your wallet balance when you exceed the daily free allowance.
What happens when credits run out?
Requests beyond the 10 daily free searches will be blocked until you buy more credits or enable auto-recharge. You will receive a 403 forbidden error with subcode insufficient_credits.
What is auto-recharge?
When enabled, Cerul automatically charges your card and adds credits when your balance drops below a threshold you set. You stay in control of the recharge amount and can configure it in your dashboard.
Why choose Pro over Pay as you go?
Pro gives you 5,000 credits per month at an effective rate of $5.98/1K versus $8/1K for pay-as-you-go top-ups. You also get higher rate limits and priority email support instead of community support.
When does enterprise make sense?
Enterprise makes sense when a team needs private indexing workflows, custom rate limits, security review, compliance review, or volume pricing. Reach out to the Cerul team to put together a plan that fits production deployments.
Can I self-host Cerul?
Yes, Cerul is fully open source. You can self-host the entire stack for free on your own infrastructure, or use the hosted API at api.cerul.ai to skip infrastructure setup and start searching immediately.
How do I get started with the API?
Sign up at cerul.ai to get your default API key in the dashboard. You receive 100 signup credits and 10 free searches per day. Then make a POST request to https://api.cerul.ai/v1/search with your Bearer token and a natural-language query.
What is included in the search response?
Each result includes a relevance score (0.0 to 1.0), tracking URL, video title, snippet, transcript (for speech segments), thumbnail URL, keyframe URL, duration, source, speaker name if detected, and timestamp start/end. When include_answer=true, an AI-generated summary grounded in matched evidence is also included.