Overview

AI proposes · humans determine · cryptography proves.

Try a company action

Normal operation

Every option uses the same deterministic gate as the real platform. Pick an ordinary read, a consequential write, or a prohibited action.

The agent is built on

Gate evaluation

no decision yet

Choose an action to see priority-ordered policy evaluation.

Open holds

Hold board

Consequential actions waiting for an IdP-verified person. Agents and API keys cannot determine holds.

Scan

Map every MCP server, tool, write path and credential. Each tool is classified by risk and by reach: how much one call can touch.

Critical findings

Coverage

Share of high-risk and critical tools that a rule holds or denies in production.

Tools and reach

tool	risk	reach	one call can touch	governed by

Shadow

Every call the agent makes is recorded and evaluated against your real policy, then allowed. Nothing holds, nothing denies. The run ends in a signed receipt you can diff against the next one.

Recorded calls

no run yet

#	tool	would	risk	reach	flags
Run Shadow to record what the agent reaches for.

receipt · none

Run diff

Run twice to see what changed between runs.

Promote to enforcement

Turn the calls Shadow would have held into hold rules. A person promotes; the change is a new policy version, ledgered under their name.

Change Gate

Deploys and infrastructure applies go through the same engine as agent tool calls. Risky changes wait for a person who is not the author; emergencies pass now and are reviewed within 24 hours.

Incoming changes

From your CI step, GitHub deployment protection rule or HCP Terraform run task.

Submit a change to see how the pack scores it.

Change history

change	env	author	rules	status

Access Gate

Privileged access is requested, determined by a person, granted for a fixed time and revoked automatically. Agents get access only through an accountable human owner. Demo clock: one minute of access lasts one second.

Requests

Choose a request.

Grants

requester	resource	status	time left

Identity register

Every agent identity needs a human owner who answers for it.

identity	owner

Connected sources

Keep your gateway; Interlok is where the proof lives. Interlok sits behind PolicyLayer and signs what reached the server, signs PolicyLayer's own records on receipt, and grades every record by what it proves.

Connect PolicyLayer

agent → PolicyLayer → Interlok observer → MCP server

No source connected.

Coverage by grade

tool	risk	G1	G2+	G2	G3
Nothing recorded from a connected source yet.

Export proof

One signed file holding a slice of your ledger, in the same format as production. Anyone can verify it offline at /verify or with npx @interlok/cli verify, without trusting Interlok.

New bundle

What to include

Content

Bundles

No bundles yet.

Share with an auditor

A link that expires, can require a one-time code sent to a named address, and is ledgered every time it is opened.

Auditor's email

Evidence Room

One signed entry per governed event, chained to the one before it and independently verifiable offline.

Signed chain

Generating P-256 key…

Independent verification

Recompute every hash, check every signature and human determination, then compare the signed Merkle anchor.

No verification run yet.

anchor · none