Decide what agents may do, and enforce it at runtime.
Keel Governance decides what agents are allowed to do and enforces it: policies, approvals, delegated-authority limits, risk controls, and revocation up to a full kill switch.
Why Keel Governance
Most AI governance tools produce registers and reports. Keel Governance is wired into token issuance, credential issuance and the Context Gateway, so a policy change or a revocation takes effect on the agent's next request.
- Who it's for
- CISOs, risk and compliance leaders and AI governance committees at companies moving agents into production on sensitive data or high-impact actions.
- Build phase
- Phase 3 · Control. See the roadmap
- First release scope
- Planned first release: policy engine with policy as code, enforcement at KeelAuth token issuance, approval workflows for onboarding and scope increases, delegated-authority limits, a kill switch across KeelAuth and KeelKey, and compliance evidence export.
What changes with Keel
| Problem | Today | With Keel |
|---|---|---|
| Policies on paper only | AI use policies live in documents nobody enforces. | Policies compiled into rules checked at token, credential and context time. |
| Unbounded agent authority | Agents can take any action their credentials allow. | Delegated-authority limits: actions, amounts, data classes, time windows. |
| No human checkpoint | High-impact actions run without review. | Approval workflows, with quorum approvals for the most sensitive. |
| Slow, partial shutdown | Stopping a misbehaving agent means hunting down every key. | One kill switch revokes identity, credentials and context access together. |
| Uneven risk treatment | Every agent is treated the same regardless of autonomy or data. | Controls tiered by Registry risk tier. |
| Manual compliance evidence | Teams assemble evidence by hand for audits and AI regulations. | Continuous evidence from enforcement decisions. |
Policies on paper only
Today · AI use policies live in documents nobody enforces.
With Keel · Policies compiled into rules checked at token, credential and context time.
Unbounded agent authority
Today · Agents can take any action their credentials allow.
With Keel · Delegated-authority limits: actions, amounts, data classes, time windows.
No human checkpoint
Today · High-impact actions run without review.
With Keel · Approval workflows, with quorum approvals for the most sensitive.
Slow, partial shutdown
Today · Stopping a misbehaving agent means hunting down every key.
With Keel · One kill switch revokes identity, credentials and context access together.
Uneven risk treatment
Today · Every agent is treated the same regardless of autonomy or data.
With Keel · Controls tiered by Registry risk tier.
Manual compliance evidence
Today · Teams assemble evidence by hand for audits and AI regulations.
With Keel · Continuous evidence from enforcement decisions.
What Keel Governance does
Policy
- One policy model covering identity, delegation, purpose, data class, tools, actions and destinations
- Policy as code with a visual editor for risk and compliance teams
- Simulation: test a policy change against recent traffic before enforcing it
Approvals and delegated authority
- Approval workflows for onboarding, scope increases and high-impact actions
- Quorum (M of N) approvals for the most sensitive operations, executed through KeelKey
- Per-agent limits: allowed actions, spending or volume caps, time windows
Risk controls
- Risk scoring per agent from Registry attributes and Intelligence signals
- Controls tiered by risk: autonomous agents on sensitive data get stricter rules by default
- Framework mappings for NIST AI RMF, ISO/IEC 42001 and the EU AI Act
Revocation and kill switch
- Revoke one token, one scope, one agent, or every agent of an owner or team
- Kill switch revokes identity, credentials and context access together, propagated through shared signals
Runtime enforcement
- The Context Gateway: an MCP-aware proxy that enforces identity, delegation, purpose, redaction and audit on every agent context request
How it works with the rest of Keel
Each Keel product feeds the others. The delegation chain and the audit trail only work when the same system issues the identity, makes the decision and records the result.
Built on standards you can evaluate
Nothing proprietary to adopt. These are the specifications the design is based on.
Put your first agent under control.
Join the design partner program: 90 days from federating your identity provider to a governed agent in production planning, with a named Keel engineer.