PromptQL is a data access agent to build AI assistants on ANY data (structured, unstructured or APIs). It gives your AI agentic data access so it can create complex query plans, run computations, retry on failures, modify query plans, and reduce hallucinations
Hi PH,
After months of hard work, trying to understand the challenges and issues around AI and data, we're excited to share PromptQL, a new data access agent from Hasura for building AI assistants to talk to your data.
You can try it out by building a GitHub issues assistant in 5 minutes here.
Although this assistant is quick to build, it is remarkably more powerful than any other assistant out there. Why? Because we have used numerous techniques to combine the power of querying structured as well as unstructured data together. Moreover, we compose data access with numeric as well as LLM computations to generate a flexible query plan to achieve complex user goals.
How much can PromptQL do? We've put together the Agentic Data Access & Computation benchmark to illustrate the various kinds of goals that users would want to achieve in their assistants and to help compare solutions.
The current way of building assistants rely on canned RAG pipelines which perform quite poorly on this benchmark, and have limited use in real-world business critical scenarios. We built PromptQL as a way to give LLMs agentic access to data so that you can build flexible and reliable assistants.
As we have been building APIs for many years now, starting as "the GraphQL on Postgres" tool and having since generalized to various data sources in our latest product, Hasura DDN, we've realized that we have the perfect foundation for crafting something optimized for AI. On top of this, we have done a lot of incredible AI research and engineering to make PromptQL.
We’d love it if you tried it out and let us know what you think!
Also, a huge thank you to @kevin for hunting us!
