The Open Dataset Every AI Developer Needs
What if the biggest bottleneck in AI agent development isn't compute or algorithms—it's simply data?
The Tool-Use Gap
I've been thinking a lot about why consumer AI agents struggle with basic tasks. The answer keeps pointing back to the same issue: we don't have quality training data for tool-use behavior.
Frontier models get this data through expensive RLHF pipelines. Open-weight models? They guess. And users suffer.
What We're Building
I'm building an open dataset specifically focused on teaching consumer LLMs to:
- Use tools reliably and verifiably
- Handle multi-step agentic workflows
- Recover gracefully from failures
- Maintain context across extended conversations
Initial focus areas:
- Code execution (sandboxed environments, debugging)
- Web interaction (forms, navigation, extraction)
- API orchestration (REST/GraphQL, auth flows)
- File operations (read, write, transform)
The 10K Trajectory Goal
We're targeting 10,000+ high-quality tool-use trajectories. But this isn't a solo project.
The best datasets emerge from diverse contributions:
- Developers sharing real workflow patterns
- Domain experts contributing examples from their fields
- Researchers defining evaluation metrics
- ML engineers running fine-tuning experiments
How to Contribute
Developers: Share your agentic workflows. What tool chains do you use? What failures do you encounter?
Domain experts: Your workflows in data analysis, research, DevOps, or content creation represent valuable training data.
Researchers: Help define what "good" tool use looks like. Your frameworks could shape how we measure success.
ML engineers: Partner on fine-tuning experiments once we have quality data.
Open Licensing, Community Governance
This dataset will be CC-BY licensed for maximum accessibility. Community governance will maintain quality over time.
The goal isn't to replicate what OpenAI or Anthropic have built. It's to create a foundational resource that anyone—researchers, startups, hobbyists—can use.
Interested in contributing? Drop a comment or reach out. Let's close the tool-use gap—together.