AI agents can investigate a case, gather information and propose an action. But when the decision is governed by business or compliance rules, those rules should not have to live inside the model.
Knowledge gives those decisions an independent policy authority. The agent gathers the facts. Knowledge evaluates the applicable rules, determines when more context is needed, returns a deterministic decision, and can issue authorization that the execution boundary can enforce.
How Knowledge works
The loop is the product. Each step is designed so the agent does what it is good at (investigating, gathering, orchestrating) while the rules stay in a governed layer the compliance function can own.
What each part of the loop does
Four capabilities materialize the loop. Each one has its own page ; this section is the map.
Progressive context - the answer to needs more context ? Knowledge tells the caller which fields the applicable rules still require, with schema and allowed values. The caller acquires them (from CRM, vendor, LLM extraction, or the user) and re-consults, until a verdict can be reached. Policies can change without changing every caller. See Progressive context.
The verdict itself - the deterministic outcome (allowed, approval_required, blocked). Produced from the encoded policy, not from LLM interpretation. Every fired rule is cited by the exact RuleVersion in force at decision time. See Verdicts and decisions for the severity ladder and precedence tie-breakers.
Enforcement - Knowledge signs every verdict as a JWS envelope binding the exact operation authorized. A Policy Enforcement Point (SDK decorator, MCP proxy, custom code) verifies the signature and checks bindings before the tool executes. An agent that hallucinates or a bug that skips the check cannot execute the underlying action, because the boundary refuses without a matching signed verdict. See Enforcement.
Auditability - every consultation writes a Consultation record freezing the exact rule versions cited, the precedence trace, the overrides applied, and the normative hash. A regulator asking "show me why this decision was made on 2026-03-15" is one API call. See Auditability.
Integrations - the surfaces through which agents, tools and existing systems connect to Knowledge : MCP proxy, Python SDK (knowledge-runtime), REST API, JWKS endpoint per tenant, webhooks, SSO / SCIM. See Integrations.
Where Knowledge fits
Knowledge adds a governed policy layer for rule-driven decisions that AI agents need to make. It works with the agent frameworks, enterprise systems and decision infrastructure you already use.
| Works with | How Knowledge fits |
|---|---|
| Your agent stack | Use LangGraph, MCP, a custom orchestrator or your existing agent platform. Knowledge provides the independent policy authority for decisions governed by business rules. |
| Your policy formalization process | Domain experts and implementation teams can bring formalized rules into Knowledge through CSV, Excel, DMN or API, then govern, test and evolve them independently of the agent. |
| Your existing decision systems | Keep FICO, existing rules engines and specialized platforms where they already work. Knowledge gives you a path for rule-governed decisions that are still handled through procedures, spreadsheets and human expertise. |
Explore Knowledge
| Build rule-governed agents | See how Knowledge fits into an agent architecture and gives implementation teams a governed place for the business rules their agents rely on. |
| Apply Knowledge to your organization | Explore how Knowledge can support policy-driven decisions across different roles, processes and regulated industries. |
| Join our Founding Design Partner Program | Work with us on one production-relevant decision, starting in shadow mode with measurable success criteria and direct access to the product team. |
| Developer documentation | Explore quickstarts, APIs, SDKs, MCP integration, security, deployment and technical reference. |
Related
| Read next | Why |
|---|---|
| Enforcement | The signed envelope and PEP model, four-actor trust chain, adoption paths |
| Progressive context | The /resolve loop, dependency inversion, why policy changes ship without touching the caller |
| Auditability | Consultation, RuleVersion, precedence trace, cold-storage replay |
| Integrations | MCP proxy, Python SDK, REST, JWKS, deployment shapes |
| Docs | Implementation-level reference |
| Solutions | Why this matters for someone in your role |