Admissibility Control · Products
Admis Security determines whether an action can be trusted. Admis Decision determines whether its consequences are acceptable. Admis Compliance determines whether its authority and obligations are satisfied. Deploy each independently or bring all three into one canonical determination.
The product portfolio
Each product controls a distinct institutional risk. They share the same Action Context, outcome model, trace identity and authoritative record, so adding a product deepens the determination without fragmenting the control architecture.
Stops compromised, anomalous, exfiltrative or insufficiently authorized actions before execution.
02 · ADMIS DECISIONEvaluates the distribution of possible outcomes when rules alone cannot resolve a material action.
03 · ADMIS COMPLIANCEApplies organizational and regulatory constraints directly to the proposed action.
This is a product portfolio, not a fixed sequence. An institution can deploy any product first. When products operate together, their evaluations contribute to one determination rather than three competing answers.
Put a Zero Trust decision at the point of action. Admis Security evaluates authentication quality, not merely authentication success, by fusing identity, delegated authority, credential posture, threat, anomaly, exfiltration and privilege evidence before a tool call can create a side effect.
auth_risk_penaltyAutonomous scrutinyAdmissibility TraceA finance agent proposes a payment to a new beneficiary using a valid but recently elevated credential.
The action is not rejected. The authority required to execute it is raised. Weak or missing evidence increases the authentication risk penalty.
Security and AI-platform teams gain one pre-action control boundary across heterogeneous agents, with the evidence and reason codes needed for operations and review.
Normalize an action once, call the Decision API through framework middleware or an MCP gateway, and translate the result through the host-native enforcement point.
Model consequence before autonomy commits. Admis Decision evaluates magnitude, uncertainty, tail risk, conditional value-at-risk, reversibility and admissible alternatives when identity and deterministic policy are not enough.
An authenticated treasury agent proposes a material allocation. No policy prohibits it, but loss is concentrated in a low-probability tail.
Risk, finance and business owners can set consequence boundaries for actions that are legitimate, compliant and still too consequential for unattended execution.
The action carries its decision class, deadline and failure policy. Fast checks stay within the host budget; deeper evaluation can pause or escalate without inventing host-specific semantics.
Turn governance intent into action-level admissibility. Governance systems determine which obligations apply. Admis Compliance determines whether this specific action satisfies its delegated authority, consent, purpose, segregation-of-duties and regulatory obligations before execution.
A service agent proposes disclosing more customer fields than the consented purpose requires. Access and purpose are valid, but the requested scope exceeds the minimum necessary.
Admis rewrites the request to the minimum consented fields and authorizes only the reduced-scope action.
Governance defines institutional intent. Admis Compliance applies it to the action, so compliance, legal and governance teams can manage versioned obligations, approvals and evidence while operations receive clear, actionable outcomes at runtime.
Developers supply the Action Context and obligation bundle identity. Admis evaluates the applicable controls and returns the same canonical outcome vocabulary used by Security and Decision.
The products together
A consequential action can be secure and compliant but still too risky. It can be low risk but outside delegated authority. The point of the portfolio is not that every product blocks the same thing. It is that every relevant reason to intervene can contribute to one authoritative answer.
The agent is authenticated, the credential is valid and the transfer is inside the formal approval policy. The transaction is still unusual, difficult to reverse and beyond the authority delegated for unattended execution.
No compromise or active threat. Credential and identity evidence are valid.
satisfiedMaterial consequence, low reversibility and concentrated downside require additional authority.
intervenesPolicy permits the transfer only after the designated approver binds to the exact action.
condition appliesTwo ways in
The products meet different buyers at the same control boundary. Enterprise teams gain governed authority across the agent estate. Developers gain a small, explicit integration contract that does not force a new model, framework or cloud.
Adopt one product or the portfolio. Prove how it would behave before granting production authority.
Begin locally with the Runtime and one integration surface. Add institutional services when the system moves into production reliance.
pip install admis.
Independent decision authority for consequential AI-agent action. Security, Decision and Compliance through one canonical core.
Choose your first control problem