Govern the AI agents
you already run
Instrument your existing agents with the KLA SDKs or OpenTelemetry (no re-platform) and KLA inserts policy-as-code checkpoints, human decision routing, and signed execution lineage at the action boundary.
Two ways to integrate
The first question evaluators ask is whether KLA forces a re-platform. It does not. Choose the control surface that matches your architecture.
Govern in place
Keep your existing agents, frameworks, and infrastructure. Drop in an SDK or emit OpenTelemetry, and KLA inserts policy-as-code checkpoints and human decision routing at the tool-call, action, and decision boundary.
- Interception happens at the execution boundary, at the point where the agent calls a tool or takes an action
- Instrument with the Node.js or Python SDK, or emit OpenTelemetry from what you already run
- No re-platform: KLA stays focused on policy, approval, and signed lineage
Run through KLA
Adopt a managed execution path when you want KLA to own more of the runtime surface: useful for greenfield Processes or teams consolidating fragmented automation.
- Tighter control surface with less local integration work
- A good fit for new Processes or replacing scattered scripts and bots
- Same policy, approval, and execution-lineage model as govern in place
What KLA works with
Govern-in-place intercepts at the execution boundary. Coverage below names what the SDKs instrument today and how everything else connects.
Agent frameworks
The SDKs instrument LangChain (Node.js and Python) and the OpenAI SDK (Node.js) today. Other frameworks connect by emitting OpenTelemetry at the action boundary. Verified governed connectivity per framework is published as it passes live conformance.
Model providers & clouds
Interception sits at the action and decision layer, not the model, so the provider behind your agent is interchangeable.
Internal systems & tools
Bring your internal APIs, databases, and workflow engines under the same policy and lineage model through the Tool Catalog: a governed inventory of tools and permissions.
Evidence & observability
KLA is OpenTelemetry-native. Every governed action produces signed Execution Lineage you can replay and hand to audit.
Governance applied at every integration
However you connect, the same controls run at the action boundary, so every governed agent earns permission to run and leaves proof behind.
Policy-as-code checkpoints
Evaluate identity, risk tier, tool access, and thresholds before an agent acts, and block over-permissioned actions before downstream systems are touched.
Human decision routing
Escalate high-stakes actions to the right reviewer in the Decision Desk, with approver identity and decision reason bound to the execution record.
Signed execution lineage
Every governed action, policy decision, and reviewer outcome is written into signed Execution Lineage you can query, replay, and use as audit-ready evidence.
Tell us your stack
Walk us through the frameworks, providers, and internal systems your agents touch, and we will map your setup to a supported integration path. Or start from the docs.