AI·rete·RAG

What it does

AI·rete·RAG is a decision-making system that combines two components. A Rete rule engine—written in pure Python—evaluates rules defined in YAML against a set of facts to produce a decision. That decision is deterministic: the same facts always yield the same verdict. A second component uses retrieval-augmented generation to explain the decision in plain English by pulling relevant passages from uploaded policy documents and citing them.

Who it is for

Organizations that must make auditable decisions in regulated domains: lending, fraud detection, clinical triage, and similar areas where decisions must be repeatable, explainable, and defensible. Teams currently using large language models for decision-making and adding guardrails afterward are the primary audience.

Pricing

The site does not show prices.

How it stands out

Most AI decision systems run a language model end-to-end and then attempt to add auditability on top. This product inverts that order. The Rete engine makes the actual decision using explicit rules, eliminating the black-box nature of pure LLM approaches. The language model is relegated to explanation only, after the decision is already made. Because the decision logic is separate from the explanation logic, the verdict is guaranteed to be deterministic—the same inputs always produce the same output—and the explanation is grounded in real policy documents rather than generated from scratch.

The use of salience-based conflict resolution in the Rete engine suggests the system can prioritize rules when they conflict, which is typical of production-grade rule engines.

What a founder should check

First, verify how mature the Rete and RAG implementation is compared to established rule engines and enterprise AI explanation tools. Rete engines have decades of history in expert systems; check whether this implementation handles the performance and scale that customers in regulated industries require.

Second, investigate switching costs for teams already running LLM-based decision systems. Organizations that have deployed and tuned language models may face friction migrating to a rule-engine-first approach, especially if their existing workflow is considered satisfactory or if rewriting rules requires domain expertise they lack.

Third, assess the moat around grounding explanations in customer documents. This is a meaningful feature for auditability, but clarify whether competitors like established rule engines with explanation plugins or wrapped LLM solutions can replicate this. Understand how difficult it is for customers to extract and maintain policy documents in a format the tool can retrieve from reliably.