What is Kite Logik?

Kite Logik is governance middleware for Python AI agents. It plugs into your existing agent framework — OpenAI, Agents SDK, LangChain, LangGraph, CrewAI, Pydantic AI, and others — and enforces deterministic OPA/Rego policy on every action the agent attempts: tool calls, sub-agent spawn, delegation, multi-step plans, resource budgets, and data classification flow.

The model still drives the agent. Kite Logik decides whether the agent gets to act on what the model proposed.

pip install kitelogik

Why this exists

Most "AI safety" today is enforced inside the model — long system prompts, output validators, content filters. That layer catches bad text. It does not catch a tool call to delete a customer record, transfer money, or shell out to a misconfigured path.

Kite Logik moves the allow / deny / require-approval decision out of the prompt and into infrastructure:

• Infrastructure enforcement. Rules are evaluated by OPA at the policy gate. The model cannot override a deny — there is no scope, role, or prompt instruction that bypasses a hard policy decision.
• Agent-level governance. Not just tool calls. Agent spawn, delegation depth, multi-step plans, resource budgets, and data classification flow are all policy-controlled.
• OPA/Rego policies. The same policy language security teams already use for Kubernetes admission control. Deterministic, testable, version-controlled, with a mature ecosystem (OPA, Conftest, Gatekeeper).
• Zero-trust sessions. Every agent gets a scoped, short-lived credential. Agents cannot expand their own permissions; child sessions can only narrow scope, never widen it.
• Immutable audit trail. Every governance decision is recorded with the policy version that decided it. SQL triggers prevent in-band tampering — the audit log is the answer to "what did the agent do, and why was it allowed?"

How it's built

Three layers, each with a single responsibility:

• Embedded SDK — @governed decorator, GovernedToolbox, framework adapters. Lives in your agent process. No network hop.
• Tether (policy engine) — talks to OPA over HTTP, or evaluates Rego in-process via Regorus. Deny-by-default, fail-closed.
• Anchor (oversight + audit) — credential broker, HITL queue, append-only audit store, OpenTelemetry tracing.

The kitelogik package ships all three. Enterprise adds a centralized Governance Gateway, a real-time dashboard, Postgres-backed storage, SSO, and compliance export packs — same policy semantics either way.

What you'll find here

• Quickstart — install, run an agent, see ALLOW / BLOCK decisions in under five minutes.
• Architecture — where the gate sits in your stack, what it intercepts, what it doesn't.
• Governance events — the structured events your policies decide on.
• Adapters — how to wire Kite Logik into the framework you're already using.