Glint-Research Archives AI Thought Paths In Fable-5-traces

Fable-5-traces is a newly released dataset that captures the behavior of the Fable 5 coding agent. This collection converts original agent logs into a format that allows users to inspect tool usage and reasoning steps. The data provides a detailed look at how an AI assistant uses terminal commands to complete coding tasks.
Glint-Research created this collection to preserve the original Fable 5 data before it disappeared. The development team converted 60 source sessions into thousands of structured training rows and trace files. They also received additional trace material from the TeichAI ecosystem to round out the final dataset.
Dataset features and training uses
- Over four thousand Pi trace sessions
- Nearly four thousand tool action records
- Detailed bash and file editing logs
- Extensive reasoning trace text data included
This resource is built for developers who want to study how AI models interact with coding tools. Researchers can use the logs to train new models on realistic reasoning and action prediction. Anyone looking to understand the step by step thought process of a coding agent will find the visualizer helpful.
Developer notes and limitations
The creators warn that this collection contains raw agent telemetry rather than clean benchmark data. Many context values are intentionally cut short so users should check the raw files for full session details. The entire dataset falls under the AGPL-3.0 license so developers must verify compatibility for commercial training.
A simple dataset of all the Fable 5 data we could get our hands on before it was taken away (no clue if it's coming back). Source: Reddit