Mahal Systems TheAgentCompany
Fits digital-worker agents that touch several internal tools and must ask coworkers for details. Borrow the self-hosted company sandbox, the LLM-played colleagues and the checkpoint-plus-result scoring.
What it asks the agent to do
- complete a work task using GitLab, a file share, a chat tool and a project tracker
- ask a simulated colleague for missing information over chat
- screen resumes or fill HR and admin spreadsheets
- write or review code and open issues
- produce a document or report and save it to the shared drive
What to borrow from it
sandbox designuser simulatorgrading designbaseline
Caveats
Skewed toward software engineering tasks; heavy Docker setup; some checkpoints are LLM-judged.
Details
| Workloads | general assistant, coding, hr, documents, workflow automation |
|---|---|
| Interaction | multi turn simulated user |
| Environment | containers |
| End state checked | Yes |
| Tools | browser, terminal, GitLab, ownCloud, RocketChat, Plane |
| Grading | programmatic check, llm judge |
| Size | 175 |
| Data | synthetic |
| Personal data | none |
| License | MIT |
| Harness | inspect_evals, own repo |
| Status | active |