Use this skill to run top-level VSS fusion search on archived video, or to ingest video files / RTSP streams for search. Do NOT use for ad-hoc visual Q&A (use vss-ask-video), live captioning (use vss-deploy-dense-captioning), or video summarization and reports (use vss-summarize-video).
The vss-search-archive skill runs NVIDIA's top-level Video Search and Summarization (VSS) fusion search across archived video, and handles ingesting new video files or RTSP streams for search indexing as well as deleting search-ingested sources. It solves the problem of finding specific events or objects inside large video archives using natural-language queries, backed by Cosmos Embed1 embeddings. It is scoped deliberately: the skill defers ad-hoc visual Q&A, live captioning, and summarization to sibling VSS skills, acting as the dedicated surface for archive search and its supporting ingestion.
Operationally it drives a deployed VSS "search" profile over REST using curl and jq, probing /docs, /health, and Elasticsearch on the host before issuing requests. It documents several discovery modes: wide-net exploratory search (high recall), narrow scoping to specific cameras and time windows, high-precision search with raised similarity thresholds and a critic for compliance-grade filtering, and metadata/tag-based filtering. Before running against a missing or wrong-profile stack it stops and asks the user to deploy via the vss-deploy-profile skill, unless explicitly pre-authorized. The skill card marks it as an alpha, demonstration-only feature not intended for production, and notes NGC rate limits, GPU memory needs, and license restrictions as limitations.
Target users are developers and engineers who need to search archived video content by natural language, index video files or RTSP camera streams, and manage those search sources. Reported evaluation results show 100% on the security dimension. Use cases include locating people, vehicles, or activities across camera archives, scoping searches to known incidents by camera and time, and PPE/compliance verification.
It runs the top-level fusion search over archived video and ingests or deletes search sources. It explicitly does not do visual Q&A (use vss-ask-video), live captioning (vss-deploy-dense-captioning), or summarization (vss-summarize-video).
An active VSS search profile reachable on $HOST_IP, with the API on port 8000 and Elasticsearch on 9200. The skill probes /docs and Elasticsearch first, and if the stack is down it offers to deploy via the vss-deploy-profile skill.
curl, jq, and Docker on the caller, plus NGC credentials in NGC_CLI_API_KEY and NVIDIA_API_KEY for any image pulls.
No. The skill card and the video search section mark it as an alpha, demonstration-only feature not recommended for production use.
Through discovery modes: wide-net mode for high recall, narrow camera/time scoping, and high-precision mode with a higher similarity threshold (e.g. 0.5+) plus the critic enabled for compliance-grade filtering.
Quick Setup:
.claude/skills/Repository
nvidia/skills