ARO-HCP IntelHub

ARO-HCP IntelHub captures releases, incidents, and documentation into a change-aware knowledge fabric. It links pull requests, deployments, and operational notes so SREs can ask what changed, why it changed, and how to respond, all through the MCP interface.

Architecture Snapshot

flowchart TD
    GH[GitHub API] -->|PR metadata| INGEST[cmd/ingest<br/>batch modes]
    INGEST -->|Fetch & checkout| CLONES[Local repo cache]
    CLONES -->|Diffs & docs| INGEST
    INGEST -->|Embeddings & summaries| DB[(internal/db<br/>Postgres + pgvector)]
    subgraph Serving
        direction TB
        DB --> MCP[cmd/mcp-server<br/>search_prs / search_docs / get_pr_details / trace_images]
        MCP --> CLIENTS[MCP clients]
    end

Loading

• cmd/ingest runs as a batch job: it pulls PR metadata from GitHub, syncs local clones, computes diffs/docs, and generates embeddings via Ollama.
• internal/db stores the precomputed metadata, embeddings, and document chunks in Postgres with pgvector for serving.
• cmd/mcp-server runs continuously, exposing search_prs, search_docs, get_pr_details, and trace_images backed entirely by precomputed content.

Local Development Workflow

1. Configure: Copy and tweak environment defaults (cp config.env config.local.env) or export variables directly.
2. Start Postgres: make compose-up launches the pgvector-enabled database via docker-compose.
3. Bootstrap Schema: make db-bootstrap (or make compose-db-bootstrap) applies migrations using cmd/dbctl.
4. Ingest Data:
   • Fast metadata only: EXECUTION_MODE=CACHE make run-ingest
   • Full pipeline: make run-ingest
5. Run MCP Server: make run-mcp starts the JSON-RPC endpoint for MCP clients.
6. Cleanup: make compose-down stops the local Postgres container when you are done.

Additional tooling: make db-status checks connectivity, make db-verify validates migrations, and make trace-images offers a CLI for image-to-source tracing.