Decionis
Start Now

Execution pattern

Gate agent memory and retrieval

Decide what an agent may read, and what it may keep.

Run a live decision — no account
Category
AI AgentsCommon in Healthcare · Fintech & payments · SaaS & platforms
Works through
langchain · salesforce · sdk · servicenow +1 more
Proof
Signed Decision DossierVerifiable against our published JWKS.
Deploy
Shadow Mode firstMeasure on your own traffic before anything is stopped.

What executes

An agent retrieves documents into context, or writes what it learned into durable memory.

What normally goes wrong

Retrieval quietly widens the blast radius of every later action. Data the actor was never entitled to read becomes data the agent can act on — and, once written to memory, data it carries into unrelated sessions.

The decision

The signals available at the moment of the action, and the rule that judges them.

Signals

  • Entitlement of the requesting actor to the source
  • Classification of the retrieved content
  • Whether the write is durable or session-scoped
  • Purpose declared for the retrieval

The gate

Restrict retrieval to what the actor is entitled to read, and refuse durable writes of classified content.

ALLOW

Retrieval proceeds within the actor's entitlement.

ESCALATE

A broader scope waits on the data owner.

BLOCK

The retrieval or the memory write does not happen.

The Protocol — Deterministic. Contextual. Explainable.Four stages left to right. Input signals — user or agent identity, context and data, tools and resources, spend and budget, policies, history and reputation, and environment — enter the Decionis Protocol. Inside, the protocol normalizes context and intent, evaluates policies and rules, runs a deterministic risk and policy engine, and explains the reason and evidence. It emits a verdict: ALLOW (authorized), ESCALATE (needs approval or more info), or BLOCK (high risk or out of policy). The verdict drives an enforcement action: allow, block, require approval, or throttle and limit. Every decision is recorded for audit, learning, and continuous improvement.INPUT SIGNALSUser / Agent IdentityContext & DataTools / ResourcesSpend / BudgetPoliciesHistory & ReputationEnvironmentDECIONIS PROTOCOLNormalizeContext & IntentEvaluatePolicies & RulesRisk & Policy Engine(Deterministic)ExplainReason & EvidenceVERDICTALLOWAuthorizedESCALATENeeds approvalor more infoBLOCKHigh risk orout of policyACTIONSAllowBlockRequire ApprovalThrottle / Limit
Four stages left to right. Input signals — user or agent identity, context and data, tools and resources, spend and budget, policies, history and reputation, and environment — enter the Decionis Protocol. Inside, the protocol normalizes context and intent, evaluates policies and rules, runs a deterministic risk and policy engine, and explains the reason and evidence. It emits a verdict: ALLOW (authorized), ESCALATE (needs approval or more info), or BLOCK (high risk or out of policy). The verdict drives an enforcement action: allow, block, require approval, or throttle and limit. Every decision is recorded for audit, learning, and continuous improvement.

No policy pack decides this pattern yet

Write the gate above as your own policy: a repository policy file, or a bundle in your workspace.

Every policy pack the evaluator ships

Prove it, then deploy it

A verdict you cannot check is an opinion, and a gate you cannot measure first is a risk of its own.

View the signed proof

Every verdict on this pattern produces a Decision Dossier: the signals it saw, the policy version that judged them, and an Ed25519 signature anyone can check against our published JWKS — without an account, and without trusting us.

See an example dossierVerify one yourself

Deploy in Shadow Mode

Run this gate against your own traffic while it stops nothing. You get the verdicts it would have returned and the actions it would have held, so the decision to enforce is made on your numbers rather than ours.

How Shadow Mode worksTry the scenario first

Related patterns

Teams that hit this one usually hit these next.

Control MCP egress

Decide what leaves before it leaves, at the tool-call boundary.

Prevent destructive production changes

Make deleting production a decision, not a command.

Prevent AI overspending

Cap what an agent can spend before the spend happens, not in next month's bill.

Related concepts

This pattern is one answer to each of these.

Agent Permissions

How do I decide what an AI agent is permitted to do?

MCP Security

How do I enforce a policy on MCP tool calls?

Policy Engine

What is a policy engine, and where does it belong?

Permission Fatigue

Why doesn't approving every AI agent action work?

Pre-Execution Authorization

Can an action be authorized before it runs rather than reviewed after?

What to do next

The same five steps for every pattern in the library.

  1. Simulate this recipe

    Run the policy against a realistic action in the browser. Push it past what the rules allow and watch the verdict come back. No account.

  2. View the signed Decision Dossier

    See exactly what was decided and why: the rule that fired, the evidence it read, the policy version in force, and an Ed25519 signature you can verify yourself.

  3. Deploy in shadow mode

    Measure what the policy would have caught on your own traffic without touching the live path. Every recipe defaults to shadow, so the first deployment carries no execution risk.

  4. Connect a real workflow

    Point the same policy at the system where the action actually originates — a checkout, an ERP posting, a Zap, an agent's tool call.

  5. Share the verified result

    Publish the proof: a public verification link, an embeddable badge, a PR comment, or an anonymized shadow-mode finding. This is how the next person discovers Decionis.

← All execution patterns