Mahal Systems GTA (General Tool Agents)
Fits assistants whose users send screenshots, photos or files and expect the right tools to be picked without being told. Borrow the human-written implicit-tool-use queries and the split between step-level tool accuracy and final-answer accuracy.
What it asks the agent to do
- answer a real question about an image, screenshot or table by chaining tools
- read handwritten or printed material with OCR and compute on it
- search the web and combine the result with a local file
- produce a deliverable from a real multi-app workflow (GTA-Workflow)
What to borrow from it
case templategrading design
Caveats
Small (229 queries) with real deployed tools that can drift; 2026 GTA-Workflow tasks are rewritten from agent platforms and forums, so some public text may be included.
Details
| Workloads | tool calling, general assistant, data analysis |
|---|---|
| Interaction | single turn |
| Environment | api sandbox |
| End state checked | No |
| Tools | image perception, OCR, search, calculator, code, drawing tools |
| Grading | programmatic check, exact match |
| Size | 229 |
| Data | real public |
| Personal data | possible |
| License | Apache-2.0 |
| Harness | own repo |
| Status | active |