A curated set of Anthropic Messages API payloads built to demonstrate what
Headroom — the context-optimization
layer for LLM apps — is good at: shrinking the large tool outputs that dominate
agentic and data-processing prompts, without changing the answer.
Each row is a complete /v1/messages request (system + tools + messages) whose
final turn is a bloated tool_result — exactly the content Headroom
compresses. Every row… See the full description on the dataset page:
https://huggingface.co/datasets/chopratejas/headroom-datasets.
