Where does Decionis sit in the stack, and how does it gate execution?

[ocularminds]

A question we've been asked by founders more than once: architecturally, where does Decionis really sit? Is it a dashboard humans check, workflow middleware, a logging layer?

[ocularminds]

At the runtime layer, consuming the execution call itself.

The high-stakes action flows through the decision. Decionis evaluates the intent against the current policy version, and:

• on allow, the wrapped function executes;
• on block, it never runs;
• on escalate, execution waits for approval.

It is not an advisory check sitting beside the code path, and not a log written after the fact — the gate is inline, between intent and effect. Every outcome, whichever branch it takes, emits a Decision Dossier.

In code, using @decionis/sdk:

import { createDecionisExecutionClient } from "@decionis/sdk";

const decionis = createDecionisExecutionClient({
  authorityBaseUrl: process.env.DECIONIS_AUTHORITY_URL,
});

await decionis.enforceAndExecute({
  request: {
    tenant_id: "tenant_demo",
    actor: { id: "research_agent", type: "AI_AGENT", runtime: "mcp" },
    action: { type: "SEND_PAYMENT", resource: "wallet.usdc", amount: 0.25, currency: "USD" },
    policy_refs: ["autonomous-spend-policy-v0.1"],
    downstream_target: { system: "payment_api", operation: "send_payment", endpoint: "POST /payments" },
  },
  execute: async ({ executionToken }) => {
    // Runs only if policy allows. The execution token binds this call to the decision,
    // so the downstream system can confirm it was authorized.
  },
});

The execute callback is the thing being governed — it is never invoked unless the decision permits it. Use enforceAndExecuteStrict when you want the execution grant redeemed before your code runs rather than after.

The same gate is reachable from several surfaces, so you don't have to hand-wire it everywhere: @decionis/langchain wraps a LangChain tool, @decionis/bedrock-guard wraps a Bedrock action group, and @decionis/mcp exposes it to MCP-speaking agents.

---

Related: What is Decionis, in plain terms? · How policies are managed and versioned · Can Decionis govern AI agent tool calls? · How Presence relates to Decionis

Have a different architecture or integration question? Start a new discussion — we answer publicly whenever we can.