Hegel AI
Hegel AI builds open-source developer tools for developing generative-AI applications. Its tools help developers evaluate prompts, models and vector databases, test for regressions, audit model safety, and monitor behavior in production.
At a Glance
- Software engineers and AI developers
- Startups building LLM applications
- Enterprise AI teams
- Researchers and teams evaluating generative-AI models
- +1 more
AI Tools by Hegel AI
(1)
PromptTools
Open Source Prompt Testing Toolkit
Discussions
No discussions yet
Be the first to start a discussion about Hegel AI
Latest News
PromptTools repository maintenance update fixed Streamlit deprecation warnings and a LanceDB import error.
O'Shaughnessy Ventures announced an investment in Hegel AI, describing PromptTools as its first platform launch.
PromptTools 0.0.45 launched Observability features and a private beta for monitoring and online evaluation.
PromptTools Playground private beta launched with hosted experiments, persistence, version control and collaboration features.
Products & Services
Open-source, self-hostable developer SDK for experimenting with and evaluating prompts, LLMs, fine-tuned models, retrieval-augmented-generation systems and vector databases; supports notebooks, code, a local playground and reusable test suites.
Hosted playground integrated with PromptTools that persists experiments with version control and provides collaboration features for teams; a hosted version is available through Streamlit Community Cloud.
Monitoring and online-evaluation capabilities integrated with the open-source library and playground, intended to observe LLM issues in real time and support reliable deployments.
Official TypeScript/JavaScript client for Hegel AI's developer platform, used to instrument and monitor LLM applications through the Hegel AI API.
Market Position
Hegel AI positions itself as an open-source, developer-focused evaluation and testing layer for LLMs, prompts and vector databases, emphasizing experimentation in familiar code/notebook workflows and a path to reusable tests and production observability. Crunchbase identifies LangChain, Braintrust and Arize AI as possible competitors.
Leadership
Founders
Kevin Tse
Investment Engineer/quantitative developer at Bridgewater Associates and software engineer at Meta (Facebook); he and Steven Krawczyk previously worked on PyTorch-related systems at Meta and Google.
Steven Krawczyk
Former engineer on PyTorch at Meta and Google, working on core library, data loading and the XLA compiler; previously worked at Amazon and Bridgewater Associates.
Executive Team
Steven Krawczyk
Co-founder & CEO
Former PyTorch engineer at Meta and Google; previously worked at Amazon and Bridgewater Associates.
Kevin Tse
Co-founder
Former software engineer at Meta and investment engineer/quantitative developer at Bridgewater Associates; worked with Krawczyk on PyTorch-related systems.
Founding Story
Kevin Tse and Steven Krawczyk met as University of Chicago undergraduates and later reconnected while working on PyTorch at Meta and Google. They started Hegel AI to make high-quality language-model evaluation and testing systems broadly available through open-source developer tools, addressing model drift, rapid model changes and the lack of continuous LLM regression testing.
Business Model
Revenue Model
Open-source PromptTools is free and self-hostable. Hegel AI has described enterprise support and early access to hosted UI and observability tools, and has reported paying customers in a private beta.
Target Markets
- Software engineers and AI developers
- Startups building LLM applications
- Enterprise AI teams
- Researchers and teams evaluating generative-AI models
- Companies using retrieval-augmented generation and vector databases
- Comparing LLMs, prompts and configurations
- Detecting model drift and GPT-4/LLM regressions
- Evaluating retrieval-augmented-generation applications and vector databases
- Auditing model safety
- Turning evaluations into reusable unit tests in CI/CD
- Monitoring deployed LLM applications
- McMaster-Carr
- Watto AI