01
CAN
Capability
Verify the agent has the technical capability to perform the proposed action.
Decision rights for autonomous AI
Capability is not permission.
Your agents can act. Mandate is being built to define when they may—with explicit authority limits, human oversight, and evidence behind each decision.
Early-stage platform · Starting with customer-service agents
Our founding commitmentsHuman-defined authorityExplicit enforcement boundariesNo invented compliance claims
01 / The missing authority layer
Four questions.
One accountable action.
A tool token tells you what an agent can access. It does not tell you what the business authorized it to do. Mandate separates capability, permission, judgment, and obligation.
Independent dimensions—not a ladder. Passing one never overrides a restriction in another.
02
MAY
Authority
Check who delegated permission, for which resource, purpose, amount, and time window.
03
SHOULD
Context
Assess business risk and prudence. Advisory judgment never creates permission.
04
MUST
Obligation
Identify binding restrictions, required evidence, logging, and escalation duties.
02 / Start with a consequential action
Same agent. Different authority.
Illustrative refund-policy examples below. These are not live evaluations or execution receipts.
Support agent / refund requestExample policy
- Delegated refund limit
- $100
- Manager approval ceiling
- $500
$45 · Eligible customer
ALLOWWithin delegated scope, with verified eligibility and required logging satisfied.
$2,500 · Same agent
BLOCKEDExceeds both limits. Route to finance under the configured policy; an ordinary manager cannot override it.
No payment service connected. No money moves in these examples.
Oversight where it matters.
Autonomy where it is authorized.
Trace the delegation
The planned registry connects every agent to an accountable owner, resource scope, and time-bound grant.
Review the exact action
Approval design binds the reviewer’s authority to the payload. A $75 approval is not a $750 authorization.
Separate decision from completion
The evidence model distinguishes proposal, evaluation, approval, execution attempt, and verified result.
Designed enforcement path
ProposeEvaluateAuthorizeExecuteAudit
Enforcement requires an integrated, server-side execution gate. A dashboard alone cannot block actions, and tools outside that boundary remain outside its control. Production integrations and security controls must be built and tested before deployment.
03 / A path from sandbox to deployment
Start with proof. Scale with confidence.
Provisional pricing hypotheses: Free Sandbox, Team around $299/month, Business around $999/month, and custom Enterprise. Scope and availability require founder approval; these are not validated commercial offers.
Free
For individuals getting started
$0/month
- 1 workspace
- Up to 3 team members
- Basic analytics
- Community support
No live paid billing is authorized by this landing page. Discuss pilot scope before making a purchase commitment.
Build with Mandate
Give autonomy a clear mandate.
Deploying support agents that issue refunds or change customer records? Help shape a design-partner pilot around your real authority boundaries.
Join design-partner early access
For CTOs, security leaders, and automation owners.