skills.rest
atomicrajat
atomicrajatCOMMUNITY
Explore all skills by this author →
GitHub

vss-deploy-video-embedding

Deploy and operate the VSS RT-Embed video embedding microservice via Docker Compose.

2|Updated Aug 20, 2026
Search Tags:#deployment#docker compose#rest api#rtsp#cosmos-embed1#nvidia vss#video embeddings
One-click install
npx skills add https://github.com/atomicrajat/industry_safety_monitoring_system --skill vss-deploy-video-embedding-atomicrajat
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: vss-deploy-video-embedding
Source: https://github.com/atomicrajat/industry_safety_monitoring_system/tree/main/.claude/skills/vss-deploy-video-embedding
Command: npx skills add https://github.com/atomicrajat/industry_safety_monitoring_system --skill vss-deploy-video-embedding-atomicrajat

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Deploying the NVIDIA VSS 3.2 GA Video Embedding (RT-Embed) microservice involves GPU prerequisites, NGC/Hugging Face credentials, a 20-minute first-boot model warmup, and a REST API with non-obvious constraints (live streams require SSE and chunk_duration > 0). This Skill gives an agent the exact Compose commands, environment variables, API call sequences, and troubleshooting steps to bring the service up and integrate it correctly. ## Core Features & Use Cases - Standalone Deployment: Bring up the rtvi-embed container with the bp_developer_search_2d Compose profile, correct env vars, and non-interactive sudo guards. - Embedding Operations: Generate text and video embeddings with Cosmos-Embed1-448p from uploaded files, HTTP/S3/file/data URLs, or live RTSP streams over SSE. - Integration & Troubleshooting: Wire Redis, Kafka, and OpenTelemetry peers, decode 400/409/422/503 error responses, and resolve model-download, permission, and healthcheck failures. - Use Case: An engineer asks an agent to start RT-Embed standalone on port 8017, wait for /v1/ready to return 200, and verify /v1/models reports cosmos-embed1-448p — the Skill supplies the full command sequence and the pitfalls to avoid. ## Quick Start Use the vss-deploy-video-embedding skill to bring up the RT-Embed service standalone on port 8017 and verify that /v1/ready returns 200 with cosmos-embed1-448p loaded.

Frequently Asked Questions about vss-deploy-video-embedding

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I deploy the VSS RT-Embed video embedding service standalone?▼

Work from deploy/docker/services/rtvi/rtvi-embed, set RTVI_EMBED_PORT, VSS_DATA_DIR, and NGC_API_KEY, then run docker compose with the bp_developer_search_2d profile. Do not use the full vss-deploy-profile workflow for standalone deployment.

How do I generate embeddings from a live RTSP stream?▼

Register the stream with POST /v1/streams/add, then call POST /v1/generate_video_embeddings with stream: true, Accept: text/event-stream, and chunk_duration greater than 0. Synchronous calls return 400 because live streams only support SSE output.

Why is /v1/ready stuck returning 503 after startup?▼

A 503 during the first 20 minutes is expected while Cosmos-Embed1 downloads and the Triton model repository builds. After that window, check for missing NGC_API_KEY, Hugging Face 429 rate limits (set HF_TOKEN), or unreachable Redis/Kafka peers.

Does the RT-Embed service require a GPU and special credentials?▼

Yes, it requires an NVIDIA GPU with the NVIDIA Container Toolkit configured as the default runtime. You also need a valid NGC_API_KEY for nvcr.io pulls and model access, and optionally an HF_TOKEN to avoid Hugging Face rate limiting.

What does a 503 'Server is busy' error on embedding endpoints mean?▼

The service processes one synchronous embedding job at a time, so concurrent requests return 503 with a busy message. Retry with exponential backoff or shard work across multiple instances.

When should I avoid docker compose down -v with this service?▼

Avoid down -v unless you want to destroy the rtvi-hf-cache, rtvi-ngc-model-cache, and rtvi-triton-model-repo named volumes. Deleting them forces a full model re-download and Triton repo rebuild, adding 20 or more minutes to the next startup.

Related Skills

Explore More Software Engineering Skills→
NVIDIA-TAONVIDIA-TAO
83

tao-run-on-docker

Run Docker containers on GPU hosts with NVIDIA Container Toolkit.

Official
Intermediate
one-ieone-ie
129

youtube-subs

Extract YouTube subtitles into clean text transcripts for LLM processing.

Official
Intermediate
berksunduriberksunduri
23

golang-observability

Instrument Go services with structured logging, Prometheus metrics, OpenTelemetry tracing, and pprof profiling.

Community
Advanced
NVIDIA-AI-BlueprintsNVIDIA-AI-Blueprints
1.8k

deploy

Automate deployment, validation, and teardown of NVIDIA VSS profiles with Docker Compose.

Official
Advanced