Generative UI and copilot stack
The GenUI stack is defined in compose-genui.yaml and contains two services:
the Chainlit copilot and Open-Intelligent-UI.
GENUI & AIOPS COPILOT STACK compose-genui SRE Operator "Diagnose CPU load & render remediation tool" Chainlit AIOps Copilot Port :8050 • chainlit-genui Step 1: pgvector RAG Lookup Embedding cosine similarity Step 2: MCP Tool Invocation Query PromQL & Loki logs Step 3: Streaming Reasoning OpenUI Lang component stream Interactive Preview iframe Embed pgvector RAG Store (PostgreSQL 17) Database: rag_storage • Table: operational_runbooks HNSW Cosine Vector Index (<=>) vector(768) • Matches nomic-embed-text Observability MCP Servers HTTP / Server-Sent Events (SSE) Grafana MCP :8000/sse Prometheus MCP :8080/sse Envoy Gateway & Host Ollama http://ollama-gateway:11435 -> 0.0.0.0:11434 32k Context Window (OLLAMA_CONTEXT_LENGTH=32768) Llama 3.2 • nomic-embed-text • Streamed Tokens Open-Intelligent-UI Canvas Port :3001 • Container: open-intelligent-ui • Next.js + @openuidev/react Container CPU Usage 92.4% Threshold Exceeded Suggested SRE Action Restart high-consumer container: ⚡ Restart Container Live PromQL Telemetry rate(container_cpu[1m])
Request flow
An operator submits a question to Chainlit on port 8050.
Chainlit asks Ollama to embed the question.
The embedding is compared with operational_runbooks in pgvector.
Chainlit sends the retrieved evidence and question to the configured Ollama model.
The streamed response appears in Chainlit, with a link to Open-Intelligent-UI when UI-oriented content is detected.
Current integration boundary
The Compose service provides Grafana and Prometheus MCP endpoint variables, but
genui/chainlit/app.py does not yet invoke MCP tools. Live telemetry retrieval
must be implemented and tested before responses can claim to include current
Grafana or Prometheus data.
See README_GENUI.md
for the deployment procedure and security notes.