The Backstory
KBAI was born out of our work on onboarding and support automation for Able CDP. We needed to create tailored setup guides for diverse customers, factoring in their business type, software stack, and setup progress. Initially, we tried using LLMs, but despite endless prompt refinement, they made unpredictable errors. It became clear that many decisions relied on implicit reasoning steps that humans naturally perform but LLMs struggled to replicate consistently. Maintaining complex prompts quickly became unsustainable.
We explored traditional non-ML solutions like Prolog and Z3, but their academic focus and complexity made them impractical for our needs. This led to the creation of KBAI - a practical, application-focused reasoning engine with an intuitive environment for managing rules without coding in obscure languages.
How It Works
KBAI converts natural language rules into a deterministic logical framework, eliminating ambiguities. Users can create, test, and refine rules through a user-friendly web interface, ensuring predictable outcomes without extensive testing. Simply show KBAI how to reason differently once, and it updates the rules from a single example.
While KBAI can function as a standalone reasoning engine—input facts, execute rules, and derive new facts—its primary strength is enhancing LLM capabilities through two integration methods:
KBAI-to-LLM: KBAI generates facts and detailed reasoning chains, which LLMs transform into clear, user-friendly responses. The full reasoning chain ensures transparency and context.
LLM-to-KBAI: When KBAI needs a fact it can’t derive (like extracting data from documents), it can pause, query an LLM for the specific information, and resume reasoning. In this mode, KBAI acts as a fully deterministic AI agent.
This hybrid approach keeps all benefits of LLMs, while ensuring predictability and making its responses detailed and self-explanatory.
Why KBAI?
Accurate Outputs, Instantly: Get precise, fact-checked responses in real-time, reducing errors and boosting user trust.
Use as an API in the cloud or download your knowledge base and engine code for offline use, giving you full control without vendor dependency.
Non-Technical Friendly: Create and modify knowledge bases from a single example, no coding required.
Try It Out
KBAI is live at https://www.usekbai.com/
The last thing to note is that the product is still in its early stages and any feedback would be greatly appreciated! Do you see KBAI fitting into your projects?