openbkn-ai/bkn-foundry

What it solves

BKN Foundry is a backend framework designed to bridge the gap between proprietary enterprise data and AI agents. It solves the problem of "context explosion" (too much irrelevant data), context decay (forgetting and hallucinations), and the lack of safe, controllable execution when agents interact with complex business logic and sensitive data.

How it works

It creates an ontology-driven "Business Knowledge Network" (BKN) that transforms scattered documents and rules into a structured network of data, logic, risks, and actions.

• Context Engineering: It uses a Context Loader to filter and rank multi-source candidates, providing agents with high-density, real-time facts and evidence citations rather than long-text concatenation.
• Harness Engineering: It implements a constraint-based execution loop (Object $\rightarrow$ Action $\rightarrow$ Rule $\rightarrow$ Constraint) that ensures tool calls are explainable, traceable, and governed by identity and access controls.
• BKN Lang: A Markdown-based modeling language that allows business experts to define knowledge networks using standard text editors without needing to write code.
• Core Components: The architecture includes VEGA for data virtualization, an Exec Factory for tool orchestration, and BKN Safe for security and auditing.

Who it’s for

It is built for enterprises and developers creating AI agents that must operate within strict business boundaries, requiring high reliability, auditability, and safe execution over proprietary data.

Highlights

• High Accuracy: Claims 93%+ accuracy in context retrieval and 15%+ improvement over baselines in reducing hallucinations.
• Cost Efficiency: Reduces token consumption by 30-50% through structured object fetching.
• Explainable AI: Provides full-chain tracing from action intent back to the original data source and active rules.
• Low-Code Modeling: Uses BKN Lang to make system rules editable as simple documents.