Early accessLLMGR-LLM-03
Agentic Tool-Use Trajectories
Human-demonstrated multi-step tasks across browsers, terminals and APIs.
- Trajectories
- 30,000+
- Environments
- 70+
- Avg. steps
- 23
Illustrative rendering of annotation layers.
01Overview
Recorded trajectories of skilled operators completing realistic tasks with tools — web browsing, spreadsheets, terminals and REST APIs — including observations, actions, intermediate reasoning and outcome verification.
Modalities
- Text
- Screenshots
- Actions
- Tool calls
Use cases
- Agent SFT
- Process reward models
- Computer-use evaluation
02Specification
Content
- Environments
- Browsers, spreadsheets, IDEs, terminals, REST APIs
- Steps
- Observation, thought, action, result
- Verification
- Outcome checked by script or second operator
03Schema
| Field | Type | Description |
|---|---|---|
| task | string | Natural-language task. |
| steps | list<Step> | Observation, thought, action, result. |
| screens | list<png> | Screenshots for GUI steps. |
| outcome | enum | success | partial | failure. |
04File structure
agentic-trajectories/ ├── data_card.md ├── trajectories.jsonl └── screens/<trajectory_id>/<step>.png
05Sample
Evaluate before you license.
- 100 trajectories across 10 environments
- Size
- ≈ 300 MB
- Formats
- JSONL, PNG
06License
Related datasets
Get started
Need more than Agentic Tool-Use Trajectories?
We extend catalog datasets with new environments, tasks, languages and annotation layers — or collect from scratch to your spec.