Platform

Runtime governance infrastructure at execution time

Complete visibility into agent workflows — observe intake, govern with runtime AI policy at dispatch, prove enforcement with signed receipts.

Observe · Govern · Prove

The default layer every irreversible agent action passes through — not session-start prompts alone.

Book 15-min proof Open Proof Lab →

Dispatch architecture — every irreversible step gated

Agent workflow

→

Intake · classify

→

Policy eval at dispatch

ALLOW→Execute + receipt

ESCALATE→Human gate→Execute + receipt

BLOCK→Fail-closed deny

Observe

Intake and classify every agent action before it enters the loop.

Policy gates

Runtime AI policy at dispatch — BLOCK · ESCALATE · ALLOW on every irreversible step.²

Human-in-loop

Approve sends, publishes, and tool calls before execution — with signed receipts.

Observe

Agent intake & classify

Tag every agent action before it enters the loop. Scope, tool use, and data-handling flags at the door.

Scope + tool classification
Data-handling flags at intake

Govern

AI policy at dispatch

Runtime policy eval on send, publish, and tool calls. BLOCK red-lines · ESCALATE to human · ALLOW with receipt.

Versioned policy packs (YAML)
100% irreversible steps gated

Prove

Receipts + replay

Signed ledger per gate · tamper-FAIL · replay demo proves policy ran before the agent acted.

6 receipt types per loop
Replay in <5 min⁸

Integrations

Mount in your stack — quickstart

Framework-agnostic dispatch layer. Generic patterns only — no vendor lock-in.

MCP tool calls

Policy eval wraps every tool invocation before execution.

HTTP webhooks

Human approve gates via webhook callback on ESCALATE.

OpenTelemetry

Trace spans for policy eval, gate, and receipt signing.

YAML policy packs

Versioned packs on disk — pinned on every signed receipt.

SIEM metadata export

Governance metadata only — not full prompt dumps.

Minimal policy pack excerpt

version: "2.4.1"
gates:
  - id: outbound_send
    on: agent.tool.external_send
    rules:
      - match: data.classification == "pii"
        verdict: BLOCK
      - match: requires_human == true
        verdict: ESCALATE
      - default: ALLOW
receipt:
  sign: ed25519
  pin_policy_version: true

Open Proof Lab → Book 15-min proof