Mahal Systems AssetOpsBench
Fits maintenance, condition-monitoring and asset-management agents in plants or utilities. Borrow its MCP-server sandbox (CouchDB work orders with Maximo field names, a read-only mode) and its comparison of tool-as-agent versus plan-executor designs.
What it asks the agent to do
- pull sensor history from an IoT store
- detect an anomaly or predict a failure mode for an asset
- create or update a maintenance work order
- coordinate several domain agents to answer one maintenance question
What to borrow from it
sandbox designtaxonomy
Caveats
Scenario count is inconsistent across the README (141-plus versus 460-plus); grading is an LLM judge (Llama-4-Maverick) over six trajectory criteria with no end-state check; the leaderboard is marked work-in-progress; IoT data is simulated.
Details
| Workloads | manufacturing, it ops, tool calling, multi agent, data analysis |
|---|---|
| Interaction | single turn |
| Environment | api sandbox |
| End state checked | No |
| Tools | MCP servers for IoT sensor data, failure modes, time series models, work orders and vibration analysis |
| Grading | llm judge |
| Size | 141 |
| Data | synthetic |
| Personal data | none |
| License | Apache-2.0 |
| Harness | own repo |
| Status | active |