ProofHound

What ProofHound is

ProofHound helps teams refine prompts for LLM-based classification. It analyzes error cases, supports prompt iteration and result validation, and provides controls to release or roll back changes. Teams can upload datasets in CSV, TSV, JSONL, JSON array, or ZIP format and map fields through the interface. Optimization can focus on overall accuracy or category-specific measures, including recall for high-risk categories and precision for error-prone classes.

The platform keeps immutable prompt versions, including their variable configurations, output rules, and differences. Experiment records cover datasets, model configurations, sample judgments, and overall and category metrics. Deployment options include gray traffic release, A/B testing, full rollout, and one-click rollback. ProofHound offers a web UI, webhooks, API tokens, and MCP connections, plus API or SDK access. Customers use their own model provider; ProofHound says it does not charge per model call. A free plan and a self-hosted open-source option are available. The product is aimed at critical classification work such as risk control, financial judgment, content moderation, and customer service intent recognition.

Who it suits

ProofHound suits operations, risk, and analyst teams working on classification tasks where category-level performance and controlled prompt releases matter. It also fits organizations that want to run the self-hosted edition with private data storage on their own infrastructure.

Limits

The free plan provides three projects, one member, three concurrent LLM calls, 5GB retained project storage, and a 200MB maximum per dataset upload. The self-hosted open-source option includes full core capabilities on one workspace deployment. Pro is planned at $29 a month, with checkout not yet open.

Getting started

1. Visit the ProofHound website.
2. Choose the free plan or the self-hosted open-source edition.
3. For self-hosting, deploy on your own infrastructure.
4. Upload a CSV, TSV, JSONL, JSON array, or ZIP dataset and map its fields.
5. Connect a model provider and use the web UI, API token, webhook, or MCP option.

Verdict

Choose ProofHound for LLM classification workflows that need prompt iteration, auditable experiments, and controlled deployment. The free plan has tight project, member, and upload limits; the Pro plan is planned but checkout is not open.