Understanding VSS Deployment Profiles: dev-profile-base, dev-profile-search, dev-profile-alerts, and dev-profile-lvs

The Video Search & Summarization (VSS) blueprint provides four modular Docker-Compose profiles—dev-profile-base, dev-profile-search, dev-profile-alerts, and dev-profile-lvs—that package distinct AI capabilities ranging from basic video understanding to live summarization and incident detection.

The NVIDIA AI Blueprint for Video Search and Summarization accelerates development by providing pre-configured VSS deployment profiles tailored to specific computer vision workflows. These profiles are self-contained Docker-Compose stacks that wire together NVIDIA NIM microservices, video analytics APIs, and supporting tools. Each profile lives in the deployments/developer-workflow/ directory of the NVIDIA-AI-Blueprints/video-search-and-summarization repository and operates as a distinct developer workflow.

What Are VSS Deployment Profiles?

VSS deployment profiles are modular, shareable Docker-Compose configurations that instantiate specific subsets of the Video Search & Summarization architecture. All four profiles share common base layers—DeepStream, the video-stream-tool (VST), and NIM inference engines—to minimize Docker footprint, but each activates different microservices and agent capabilities through profile-specific environment variables and YAML configurations.

The Four VSS Deployment Profiles Explained

dev-profile-base

The dev-profile-base profile provides general video-understanding and report generation capabilities, serving as the default "hello-world" stack for the blueprint.

• Core Services: vss-agent, video-analytics-api, vst (video-stream-tool), and optional vss-search-analytics
• Primary Use Case: Basic video ingestion, metadata extraction, and summary report generation without specialized downstream consumers
• Key Files:
  • Environment variables: deployments/developer-workflow/dev-profile-base/.env
  • Agent configuration: deployments/developer-workflow/dev-profile-base/vss-agent/configs/config.yml

dev-profile-search

The dev-profile-search profile enables an end-to-end search-and-RAG (Retrieval-Augmented Generation) workflow, allowing developers to ingest video, generate embeddings, and query content using natural language.

• Core Services: Base stack plus vss-search-service (RAG backend) and vss-search-analytics (2-Dimensional search UI)
• Primary Use Case: Semantic search across video libraries with thumbnail-rich result interfaces
• Key Files:
  • Environment variables: deployments/developer-workflow/dev-profile-search/.env
  • Agent configuration: deployments/developer-workflow/dev-profile-search/vss-agent/configs/config.yml
  • Analytics config: deployments/developer-workflow/dev-profile-search/video-analytics-2d-app/vss-search-analytics-$STREAM_TYPE-config.json

dev-profile-alerts

The dev-profile-alerts profile specializes in incident analysis and automated alert generation for scenarios such as traffic violations or safety compliance monitoring.

• Core Services: Base stack plus vss-video-analytics-api and vlm-as-verifier (Vision-Language Model alert verifier)
• Primary Use Case: Real-time incident detection with alert payload generation for downstream consumers
• Key Files:
  • Environment variables: deployments/developer-workflow/dev-profile-alerts/.env
  • Agent configuration: deployments/developer-workflow/dev-profile-alerts/vss-agent/configs/config.yml
  • Alert model config: deployments/developer-workflow/dev-profile-alerts/vlm-as-verifier/configs/alert_type_config.json

dev-profile-lvs

The dev-profile-lvs (Live Video Summarization) profile delivers real-time inference capabilities with streaming visualization for continuous video streams.

• Core Services: Base stack plus vss-lvs-service (live summarization micro-service) and vss-lvs-ui (Kibana dashboard)
• Primary Use Case: Near real-time video summarization with streaming dashboards for operational monitoring
• Key Files:
  • Environment variables: deployments/developer-workflow/dev-profile-lvs/.env
  • Agent configuration: deployments/developer-workflow/dev-profile-lvs/vss-agent/configs/config.yml
  • LVS service configs: deployments/developer-workflow/dev-profile-lvs/vss-lvs-service/configs/*.yml

How the Profiles Are Wired Together

The orchestration of VSS deployment profiles relies on Docker Compose's include directive and environment variable injection.

Top-Level Composition: The deployments/developer-workflow/compose.yml file aggregates all four profiles via path-based includes:

# deployments/developer-workflow/compose.yml

services:
  # Shared infrastructure definitions

includes:
  - path: ./dev-profile-base/compose.yml
  - path: ./dev-profile-search/compose.yml
  - path: ./dev-profile-alerts/compose.yml
  - path: ./dev-profile-lvs/compose.yml

Environment Variable Injection: Each profile exports VSS_AGENT_CONFIG_FILE and other critical paths through its dedicated .env file. When you source a profile's environment, Docker Compose receives the variables required to mount the correct agent configuration and model directories.

Agent Configuration: The vss-agent/configs/config.yml file serves as the single source of truth for pipeline orchestration. It defines video source ingestion endpoints, inference model loading (NIM, DeepStream), and downstream tool behavior (e.g., video-report-gen). Profile differentiation occurs through YAML section toggling—enabling search embeddings, alert verifiers, or LVS modules based on the mounted configuration.

Deploying and Switching Between Profiles

Starting a Specific Profile

To launch a particular VSS deployment profile, source its environment file before executing Docker Compose:

# From repository root - activate search profile

source deployments/developer-workflow/dev-profile-search/.env

# Launch only the services defined in dev-profile-search and shared base

docker compose -f deployments/developer-workflow/compose.yml up -d

The source command exports VSS_AGENT_CONFIG_FILE, model paths, and data directories required by the compose stack.

Switching Profiles

To migrate between workflows without image rebuilds:

# Terminate current stack

docker compose -f deployments/developer-workflow/compose.yml down

# Activate alerts profile

source deployments/developer-workflow/dev-profile-alerts/.env
docker compose -f deployments/developer-workflow/compose.yml up -d

Running the Agent Manually

For debugging or development outside the containerized orchestration:

nat serve \
  --config_file $(pwd)/deployments/developer-workflow/dev-profile-base/vss-agent/configs/config.yml

The nat binary is the NVIDIA AI Toolkit CLI bundled in the repository.

Inspecting Effective Configuration

To verify which modules are active for a given profile:

# Substitute environment variables and parse the YAML

envsubst < deployments/developer-workflow/dev-profile-base/vss-agent/configs/config.yml | yq .

This displays the merged configuration after variable substitution, revealing enabled search, alerts, or LVS sections.

Key Configuration Files

Category	File Path	Purpose
Profile Aggregation	deployments/developer-workflow/compose.yml	Top-level Docker-Compose including all four profile definitions
Base Profile	deployments/developer-workflow/dev-profile-base/compose.yml	Service graph for general video understanding
Search Profile	deployments/developer-workflow/dev-profile-search/compose.yml	RAG and search-analytics service definitions
Alerts Profile	deployments/developer-workflow/dev-profile-alerts/compose.yml	Incident detection and verifier service wiring
LVS Profile	deployments/developer-workflow/dev-profile-lvs/compose.yml	Live summarization and Kibana UI services
Alert Model Config	dev-profile-alerts/vlm-as-verifier/configs/alert_type_config.json	VLM prompt templates and alert classification rules
Search Analytics Config	dev-profile-search/video-analytics-2d-app/vss-search-analytics-$STREAM_TYPE-config.json	2D UI rendering and embedding visualization parameters

Summary

• VSS deployment profiles provide four specialized Docker-Compose stacks—dev-profile-base, dev-profile-search, dev-profile-alerts, and dev-profile-lvs—each optimized for distinct video AI workflows.
• All profiles share base infrastructure (DeepStream, VST, NIM) but diverge in microservice activation through profile-specific .env files and agent YAML configurations.
• Profile switching requires only sourcing the target environment file and restarting the top-level compose.yml, eliminating container rebuild overhead.
• Configuration management centers on VSS_AGENT_CONFIG_FILE variables pointing to vss-agent/configs/config.yml within each profile directory.

Frequently Asked Questions

What is the difference between dev-profile-base and dev-profile-search?

dev-profile-base provides fundamental video ingestion and report generation without persistent vector search capabilities, while dev-profile-search adds the vss-search-service for RAG operations and a 2-Dimensional analytics UI for querying video embeddings. The search profile mounts additional configuration files defining embedding models and vector database connections in dev-profile-search/video-analytics-2d-app/.

How do I switch between VSS deployment profiles without rebuilding containers?

Execute docker compose -f deployments/developer-workflow/compose.yml down to stop the current stack, then source deployments/developer-workflow/<profile-name>/.env to load the target profile's environment variables, and finally docker compose -f deployments/developer-workflow/compose.yml up -d. This workflow reuses existing image layers while remounting the correct agent configuration and service definitions.

Which profile should I use for real-time video analysis?

Use dev-profile-lvs (Live Video Summarization) for real-time workflows, as it includes the vss-lvs-service microservice and Kibana dashboard (vss-lvs-ui) designed for continuous stream processing and near real-time summarization. For incident detection rather than summarization, use dev-profile-alerts, which activates the vlm-as-verifier for immediate alert generation.

Where are the model configurations stored for each profile?

Each profile stores model configurations within its directory subtree under deployments/developer-workflow/. For example, the alerts profile stores VLM verifier settings at dev-profile-alerts/vlm-as-verifier/configs/alert_type_config.json, while the search profile maintains analytics configurations at dev-profile-search/video-analytics-2d-app/vss-search-analytics-$STREAM_TYPE-config.json. The agent locates these files via paths defined in the profile's .env file.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}