AI Video Search: Natural Language Across Your Entire Library
VidForgeX indexes every analysis result as vector embeddings, enabling semantic search across your entire video library. Ask questions in plain language and receive timestamped, source-attributed answers from your footage.
From video archive to queryable knowledge base
Most organizations have years of recorded video that is effectively inaccessible — stored but unsearchable. VidForgeX transforms every analyzed video from an opaque media file into a row in a queryable knowledge base. The transformation happens in three steps:
- Analysis — AI processes the video and produces a structured JSON artifact covering events, persons, metrics, and insights
- Chunking — The artifact is split into semantic segments: individual activity sequences, metric summaries, insight paragraphs, and person sighting events
- Embedding — Each chunk is embedded into a high-dimensional vector and stored in a vector index alongside source metadata (video ID, timestamp, segment type)
At search time, your query is embedded in the same space and matched against all indexed chunks via cosine similarity. The top matches are returned to the AI model as context for generating a grounded, attributed answer.
Search accuracy: retrieval vs generation
VidForgeX uses a RAG (Retrieval-Augmented Generation) architecture. The retrieval step (vector search) applies a similarity threshold — chunks below this threshold are excluded, reducing hallucination risk. If insufficient relevant chunks are found, VidForgeX falls back to direct artifact retrieval rather than generating an answer from model knowledge alone.
Every answer includes attribution: which video(s) the information came from. This allows users to verify answers by navigating directly to the source video and timestamp.
Multi-video and organization-wide search
Search scope is user-controlled per query:
- Single video — search within one specific recording (1 AI credit)
- Folder — search all videos in a folder (2 AI credits)
- Organization — search across entire org library (3 AI credits)
- Person library — search all videos tagged to a specific person (auto-resolves up to 200 videos)