rhdh-developer-lightspeed

name	rhdh-developer-lightspeed
metadata	{"author":"rhoai3-coding-demo","version":"1.0.0","platform-family":"rhdh","platform-baseline":"repo","ocp-baseline":"repo","skill-group":"Developer Hub"}
description	Use when integrating Developer Lightspeed AI assistant with Red Hat Developer Hub for AI-driven developer portal interaction. Covers LCORE sidecar architecture, LLM provider configuration (vLLM, OpenAI, Ollama, Vertex AI), RAG embeddings, safety guards, chat history, user feedback, Notebooks, and air-gapped mirroring. Do NOT use for MCP tool configuration (use rhdh-mcp-tools). Do NOT use for RHDH installation (use rhdh-install-*).

RHDH Developer Lightspeed

Configure Red Hat Developer Lightspeed for Red Hat Developer Hub 1.10 — an AI-powered virtual assistant providing chat assistance, RAG-grounded responses, Notebooks for private research, and model evaluation capabilities.

When to Use

• Configuring LLM provider connections (vLLM, OpenAI, Ollama, Vertex AI)
• Setting up Developer Lightspeed via Operator or Helm chart
• Configuring RBAC for lightspeed.chat.read / lightspeed.chat.create
• Enabling/disabling the Lightspeed floating action button (FAB)
• Configuring RAG embeddings and vector database
• Setting up safety guards (Llama Guard)
• Enabling user feedback collection and chat history persistence
• Configuring Notebooks (Developer Preview)
• Mirroring images for air-gapped environments
• Customizing system prompts for LLM behavior

Architecture

• LCORE (Lightspeed Core Service): Sidecar container managing LLM interactions, MCP tool runtime, safety providers, vector DB, chat history
• RAG init container: Copies embedding data to shared volume
• FAB interface: Floating action button on all RHDH pages
• BYOM (Bring Your Own Model): No bundled model; connect any OpenAI-compatible inference provider

Supported LLM Providers

Provider	Env Variable	Notes
vLLM	ENABLE_VLLM	Append /v1 to URL manually; used for RHOAI
OpenAI	ENABLE_OPENAI	Cloud-based inference
Ollama	ENABLE_OLLAMA	Desktop/cluster inference
Vertex AI	ENABLE_VERTEX_AI	Google Cloud Gemini models

Important: Set ENABLE_* variables to "true" to activate. Leaving unset disables. Setting to "false" does NOT disable (system checks if defined).

Prerequisites

• RHDH 1.10 deployed (Operator or Helm)
• Active API key/credentials for chosen LLM provider
• Cluster administrator privileges

Validation

# Verify LCORE sidecar running
oc get pods -l app=backstage -o jsonpath='{.items[*].spec.containers[*].name}'

# Verify FAB appears (functional check in browser)
# Check RBAC permissions
oc logs deployment/<rhdh> -c backstage-backend | grep lightspeed

Boundaries

• Lightspeed is enabled by default in RHDH 1.10
• Notebooks are Developer Preview — not production-ready
• Model evaluation framework is Developer Preview
• Red Hat does not collect or access feedback/chat data
• MCP server configuration within LCORE is covered by rhdh-mcp-tools skill
• Vertex AI has limited testing and custom architecture mapping

References

• references/source-capture.md — source ledger
• references/official-doc-extraction.md — extracted procedures and config