ProofHound Docs
Optimize and operate LLM classification prompts on hosted ProofHound — from a labeled dataset to a monitored, reversible production release.
ProofHound turns a labeled dataset and a task description into a measurably better classification prompt — running the loop of error analysis, rewrite, experiment, and comparison for you — then manages that prompt's whole life: version history, experiments, connectors, and a canary-tested, reversible production release. These docs cover the hosted platform; you sign in and start working, with no setup.
Start here
Why ProofHound
The idea behind the product: classification is moving to LLMs, the real work is iterating the prompt, and that work should be automated and managed like production software.
Run your first optimization
Turn a small labeled dataset into a measurably better prompt end to end — about 15 minutes, no code required.
Self-host ProofHound
Prefer to run the open-source build yourself? Bring up the whole stack with Docker Compose.
Find your way around
The docs are organized into four kinds of page, so you can tell learning from doing from looking up. Pick by what you need right now.
Tutorials
Learning-oriented lessons that take you by the hand from nothing to a first working result.
How-to guides
Problem-oriented recipes for getting a specific task done — import a dataset, connect a model, publish a prompt.
Reference
Precise, look-up-oriented facts: dataset formats, model configuration, optimization settings, metrics, roles, and the API.
Explanation
Understanding-oriented discussions of why the optimization loop, the metrics, and the object model are shaped the way they are.