Platform Overview
Platform Overview
Architecture
Components
REST API
All interaction happens through the REST API at https://api.pureframe.ai. The API handles authentication, rate limiting, quota checks, and routing to the appropriate backend service.
Processing pipeline
When you upload a video, a background job runs through these stages:
- Frame extraction — frames are extracted at regular intervals
- Visual embedding — each frame is encoded into a dense vector representing its visual content
- Speech transcription — audio is transcribed and aligned to frame timestamps
- Indexing — vectors and transcripts are written to the vector store, ready for search
The pipeline is asynchronous. You poll GET /v1/jobs/{job_id} to track progress.
Vector store
Search is powered by a vector database. When you submit a search query, Pureframe encodes it using the same method as your video frames and finds the closest matches — this is what makes natural language search work without exact keyword matching.
The vector store holds one entry per indexed frame and speech segment, each tagged with your user_id and collection_id for isolation.
Object storage
Video files and extracted frame images are stored in cloud object storage. All URLs returned by the API are presigned — they expire after ~1 hour. Store video_id values, not URLs.
MCP server
The MCP server at https://mcp.pureframe.ai wraps the agent tools (search_videos, list_collections, get_collection, get_video) in the Model Context Protocol so AI clients can call them directly.
Data isolation
Every piece of data in Pureframe is scoped to your account. API keys only have access to your own collections and videos — there is no cross-account data visibility.
Availability
The API and processing pipeline are hosted on cloud infrastructure with automatic scaling. Rate limits protect against individual key abuse; see Rate Limits for the per-endpoint limits.
