BMC-Sentinel

BMC-Sentinel is an openUBMC-oriented intelligent operations POC for Track 1, “Rebuilding BMC: AI-era intelligent O&M innovation platform”.

The project is intentionally not a generic chatbot. It exposes BMC telemetry, logs, alarms and guarded control actions as Model Context Protocol style tools, then uses a lightweight multi-agent pipeline to collect evidence, diagnose root cause, enforce safety policy and generate an auditable report.

What Is Delivered

• ubmc_sentinel/: Python MCP stdio server, tool registry, adapters, safety policy, audit logger, diagnosis engine, predictive maintenance, cooling optimizer and dry-run self-healing runbook engine.
• aiops_adapter/: openUBMC component POC, including MDS model, Lua component source, service file and manifest snippet.
• rag_knowledge_base/: lightweight rule and knowledge files for thermal, power and safety decisions.
• demo_cases/: repeatable demo scenarios for judging and presentation.
• docs/: design document, integration guide, API reference, security model and test report. docs/competition_mapping.md maps the work to the Track 1 scoring requirements.
• tests/: runnable unit tests for functional, safety and protocol behavior.

Quick Start

Run the unit tests:

python -m unittest discover -s tests -v

Run a natural-language demo:

python -m ubmc_sentinel.server --scenario thermal_risk --demo "这台服务器风扇为什么一直高转？"

Start the MCP stdio server:

python -m ubmc_sentinel.server --stdio

Call one tool without an MCP client:

python -m ubmc_sentinel.server --scenario thermal_risk --tool diagnose_fault

Run the new high-score POC tools:

python -m ubmc_sentinel.server --scenario thermal_risk --tool predict_maintenance_risk
python -m ubmc_sentinel.server --scenario normal --tool optimize_cooling_power
python -m ubmc_sentinel.server --scenario thermal_risk --tool generate_self_healing_plan

openUBMC Integration

The POC follows a two-layer integration strategy:

1. ubmc_sentinel can run beside an openUBMC QEMU or real BMC instance and read resources through SSH, CLI, IPMI, D-Bus style commands or Redfish.
2. aiops_adapter is an openUBMC application component skeleton that stores AI diagnosis state, safety policy and audit summaries inside openUBMC resource models.

See docs/integration_guide.md for the exact integration and build route. See docs/competition_mapping.md for the competition requirement checklist.

Safety First

Read-only tools execute automatically. Risky controls such as fan policy change and power control require confirmation and default to dry-run. Every tool call is recorded in logs/audit.jsonl with sensitive parameters masked.