Percival Integrations

June 30, 2025

Introduction

We know that agent debugging is hard with content explosion, long traces, and many points of failure. Traditional evaluation methods simply don’t work. They are static and measure the performance of a singular checkpoint, failing to capture the dynamism of agentic workflows.

Percival is a SOTA AI agent debugger, and it’s here to help. It is built to robustly analyze agent traces to catch 20+ failure modes, including reasoning, planning, and execution.

Companies like Nova AI have utilized Percival to boost productivity 60x by reducing agent debugging time from 1 hour to 1 minute, fix 3 agent failures in 1 week with automated prompt suggestions, and increase agent accuracy by 60% on the SAP tool dataset through experimentation. You can read more about their usage of Percival here.

Integrations

To make it easier for developers to begin using Percival, we’ve integrated this with a few popular development platforms, so you can experience Percival within the comfort of your existing tooling ecosystem.

Smolagents - a simple library that unlocks agentic capabilities for language models.
OpenAI Agents - a rich set of composable primitives that enable you to build agents.
Pydantic - a Python agent framework designed to build production-grade applications.
CrewAI - a lean, lightning-fast Python framework built entirely from scratch.
Langchain - a platform for controlled, custom agent workflows and reliability support.

Conclusion

Agent evaluation is complex, dynamic, and time-consuming. Let Percival help you rigorously analyze your traces, so you can spend more time building what’s next. We’re excited to see what your team unlocks with the power of Percival! 

Reach out to us with any questions or concerns.

Thank you! Your submission has been received, we'll be in touch soon!
Oops! Something went wrong while submitting the form. Please try again.