Patronus AI
Hallucination and safety evaluators (Lynx) — now pivoting to agent simulation.
About
Patronus AI built its name on automated evaluators for hallucination and safety — notably Lynx, a fine-tuned groundedness evaluator — plus rubric scoring, experiments, and guardrails, with Python/TypeScript SDKs.
Strategic note (June 2026, TechCrunch): Patronus raised a $50M Series B and launched 'Digital World Models' — simulation environments for agent training/testing — signaling a pivot away from pure evaluation. The eval platform's long-term future is unclear; evaluate accordingly before building on it.
What it can do
- Lynx hallucination evaluator
- Rubric-based and safety evaluators
- Experiments and A/B comparisons
- Python/TypeScript SDKs
- Digital World Models (agent simulation, new)
Setup
- Sign up at patronus.ai and create an API key.
- Install the SDK (
pip install patronus). - Call the /v1/evaluate endpoint with built-in or custom evaluators.
- Note the strategic pivot before committing to a long-term build.
Prerequisites
Patronus account; API key.
Use cases
Works with
Trust score — 6/10
Scored 6/10 (metrics tier). Real evaluation technology (Lynx) with research backing, but the June 2026 pivot to agent simulation puts the eval platform's future in question — a material caveat. Closed SaaS. Metrics tier with the pivot flagged prominently.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 7.0/10
- Maintenance · recency, releases, issue load
- 5.0/10
- Openness · license + docs
- 4.5/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · provider-tier estimate — thinner evidence (no public adoption metric)
- · $50M Series B Jun 2026 (TechCrunch); pivoting to agent simulation — provider-tier estimate
- · no release date — neutral maintenance score
- · manual review: Lynx hallucination eval verified; strategic pivot to Digital World Models means the eval platform's future is unclear; license: closed SaaS
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.