MetaMynd User Guide

The gate's possible answers

Allowed

The gate permitted this request under the agent's mandate and rules. Permitted is not the same as executed — see the execution stage.

Observe

Permitted, but a rule flagged it for attention. The action could go ahead; the flag is recorded for review.

Escalate

Held for a person to approve, deny or modify. Nothing runs until someone decides, and the review expires if nobody does.

Block

Refused. The request broke the mandate or a rule, so nothing was authorised to run.

Suspend

Refused, and the agent is now suspended — it is denied every action until an owner reinstates it.

Quarantine

Refused, and the agent is now quarantined — isolated and denied every action pending investigation.

Decommission

Refused, and the agent is permanently retired. A decommissioned agent cannot be reinstated.

Create an account

1. Open the site and choose Sign up (or go to /auth/signup).
2. Fill in First Name, Last Name, Email, Contact Number (at least 10 characters, e.g. +60123456789), Gender, Password and Confirm Password.
3. Click Sign up.
4. Open the confirmation email and click its link (it expires after 24 hours), then Go to sign in.

Password rules

• At least 12 characters. No other complexity rules.
• Passwords found in known data breaches are refused ("This password has appeared in a known data breach").

Sign in

1. Go to /auth/login, enter Email and Password, click Log in.
2. If two-factor authentication is on, enter the 6-digit Authentication code from your app and click Verify. You have 5 minutes.

Forgot your password

1. On the login page click Forgot your password?, enter your email and click Send reset link.
2. The email link works once and expires after 1 hour. Choose a new password (12+ characters) and click Reset password.

Joining an existing organisation

If a colleague invited you, use the link in their invite email instead of the normal sign-up: your email is filled in, you choose a password and click Accept invite and sign up, and you can sign in straight away. See §4.

Turn on two-factor authentication (recommended)

1. Go to Settings → Security → Two-Factor Authentication and click Enable Two-Factor Authentication.
2. Scan the QR code with an authenticator app (Google Authenticator, 1Password, Authy…). If you cannot scan, use Can't scan it? Enter this key manually.
3. Type the 6-digit code and click Confirm and enable.
4. Ten backup codes appear. Click Copy codes and store them safely before clicking Done: they are never shown again. Each works once, in place of an app code.

To turn it off: enter your current password under Confirm your current password to disable 2FA and click Disable 2FA.

Watch out for

• One session per account. Signing in somewhere else — another browser, or the create-metamynd-agent command-line tool — signs out the older session.
• "Confirm your email address before signing in" means the confirmation link was not used. Click Send me a new link.
• Rate limits: repeated sign-in, sign-up or reset attempts from one network are paused for a few minutes. The message tells you how long; your password was not checked.
• Sessions refresh automatically; after about a week of inactivity you are asked to sign in again.

What each tab is for

Tab	Use it to
Profile	Edit your First Name, Last Name and Phone Number and click Save Changes (email cannot be changed). See your identity-verification badge; Complete Verification takes you to KYC / KYB. The Danger Zone closes your account (below).
Security	Change your password (Update Password), turn two-factor authentication on or off, and set Auto-Lock.
Governance	Arm, trip and reset circuit breakers, and see which platform-wide enforcement switches are on.
Wallet	See your Hedera account, DID document, balance (tℏ on testnet) and your testnet/mainnet network identities. Read-only.
Billing	See Mainnet write usage: the free tier includes 100 mainnet writes (identity mints and verification anchors). Testnet is unlimited.
Notifications	Turn the Weekly refusal digest (email) on or off.
Preferences	Set the Reporting timezone, Dark Mode, Language, Date Format, Font Size, High Contrast Mode, Reduce Motion and Usage Analytics.

Your preferences are saved to your account, so they follow you to another browser or device. Each switch applies straight away.

Display and language

• Language switches the site between English, Spanish, Traditional Chinese and Japanese. Some dashboard pages are still in English only.
• Date Format changes how dates are shown on the pages that use the shared date display (for example the Activity Log, incidents, agent profiles and Regulator Evidence).
• Font Size, High Contrast Mode and Reduce Motion apply across the dashboard.
• Turning Usage Analytics off stops usage tracking for your account.

Auto-lock

Under Security → Auto-Lock, turn it on and choose the Auto-Lock Time (minutes). If you are inactive for that long — in every dashboard tab — you are signed out and need to sign in again.

Close your account

Closing an account stops everything it governs but keeps its records. Evidence, decisions, the audit log and your agents' history stay intact, because regulators and counterparties may rely on them.

1. Go to Settings → Profile → Danger Zone and click Close account.
2. Enter your Current password, type DELETE in Type DELETE to confirm, and click Close account.

MetaMynd then suspends every agent you own, revokes all your active mandates, signs you out everywhere and replaces your name and phone number with placeholders. You can no longer sign in; the login page says the account is closed. If something cannot be stopped, the account stays open and the dialog tells you, so you can try again.

Set the reporting timezone

Reporting pages and the audit log group days and show times in this zone. Records are always stored in UTC, so changing it never changes data — only how it is displayed and which calendar day an event falls into.

1. Open Settings → Preferences → Language & Region.
2. In Reporting timezone, choose a zone by its region name (for example Asia/Kuala_Lumpur or Europe/London), or Not set — use each viewer's own timezone.
3. Click Save timezone. The reporting pages then say "Timezone: … (your organisation's setting)".

On any reporting page you can still pick a different zone for that one view from the Timezone filter. The change is recorded in the audit log (tenant.timezone_set).

Watch out for

• Offsets such as UTC+8 are rejected; use a region name such as Asia/Singapore.
• A phone number already used by another account is refused when you save your profile.
• There is no API-key tab: agents authenticate with their own keys and config file, issued by the Launchpad.

Before you start

• For an organisation: its legal name, registration number and registry jurisdiction (two-letter code, e.g. MY).
• Somewhere safe to store a private key.

Submit a principal

1. Choose Individual (KYC) or Organization (KYB).
2. Enter a Display name. For an individual, the Full legal name is optional (stored encrypted). For an organisation, Legal name, Registration number and Registry jurisdiction are required.
3. Click Submit for verification.
4. A green box shows the principal's private key, once. Reveal it, copy it and store it safely, together with the public key and case reference.
5. Complete the identity check. Depending on how your deployment is set up, either click Complete verification / verify now to open the hosted identity check (ID document and selfie, or company documents), or wait for a MetaMynd reviewer ("awaiting review").

Testnet and mainnet

Testnet

Hedera's test network — no real money, for building and trying things.

Mainnet

Hedera's live network — real value. Needs a verified accountable entity.

• During the beta, a new organisation's first principal may be self-approved automatically. A self-approved principal can deploy testnet agents only; no identity documents were checked.
• To go to mainnet, use the Verify for mainnet card on this page: choose Identity check (Start identity check) or Manual review (Request manual review). It stays the same principal; nothing needs re-creating.
• Only a principal verified by the identity provider, or by a person at MetaMynd, unlocks mainnet.

What the statuses mean

Unverified

No identity check has been done. Testnet only.

Pending

The identity check has been submitted and is waiting for a result.

Finalising

The check passed and the verified identity is being issued.

Verified

The identity check (KYC for a person, KYB for an organisation) passed. Needed for mainnet.

Self-approved

You approved this yourself, with no identity check. Allowed on testnet only.

Renewing

A previous verification is being renewed.

Rejected

The identity check failed.

Expired

The verification lapsed and must be renewed.

Under each status a basis line says how it was verified: "self-approved · no identity check · testnet only", "reviewed by MetaMynd compliance", or "verified by {provider}".

Watch out for

• A verification expires. Expiry dates within 14 days turn amber ("Expiring soon — renew to keep SOPs active"); use renew before it lapses.
• The agent DID for a principal is minted only after verification succeeds.
• Submissions are limited to 10 per hour.

Before you start

Members and invites work right away, before you have a legal entity — a new organisation admin can invite people on day one. The other tabs are about one legal entity: until you submit a principal under KYC / KYB, the page says "No legal entities yet", Overview, Groupings, Agents and SOPs each show a Go to KYC / KYB button, and Supervision explains that access is granted per legal entity. Management controls (invites, role changes, adding units) are shown to organisation admins; a member sees the Members list read-only.

Tabs

Tab	What you do there
Overview	Check the legal name, registration number, jurisdiction, principal ID and DID, the verification badge and assurance level, and the on-chain proof ("verify on HashScan"). Network identities (testnet/mainnet) are listed below.
Groupings	Create operating units — e.g. "Payments Desk" — with Unit name and Add unit. Units appear in Governance Overview → Control health by operating unit and in the Agent Register.
Agents	See the entity's governed agents and assign each to a unit. Adopt unassigned agents attaches any of your agents that have no entity yet.
SOPs	Opens the SOPs (Rules) page for this entity.
Members	See everyone in your account and invite new people (below). An admin can change another member's role between Organization Admin and Member; you cannot change your own role.
Supervision	Grant, review and revoke a regulator's read access. See §31.

Invite a member

1. On the Members tab, under Invite member, enter their email, choose the role (Member or Organization Admin) and click Send invite.
2. They receive an email with a link (valid 7 days). Opening it shows "You're joining … as …" with their email filled in; they choose a password and click Accept invite and sign up. No separate confirmation email is needed.
3. Track invites in the Invites list. Resend sends a fresh link (the old one stops working); Revoke cancels an invite.

Members join your organisation and can be named as approvers for held actions. They do not see your agents, mandates or rules — each person's records stay their own.

Watch out for

• An invite works only for the email it was sent to, and only once. An address that already has a MetaMynd account cannot be invited.
• A registration number beginning BETA- is a beta placeholder, not a real registry record.

The Getting started checklist

1. Verify your identity — submit a principal (§3). While it is pending you can still Try the sandbox agent.
2. Create your agent — Create agent opens the Launchpad. If your identity is still in review, Create agent (testnet) lets you start on testnet now and keep the same agent after verification.
3. Set its rules — done once an agent has a mandate and at least one active SOP. Set mandate and rules or Add an SOP take you to the right place.

Two extra rows may appear: Go live on mainnet (optional) and Restrict who can claim your holds — shown when an agent can move money and any service could still execute its approved actions (see §16). Step states read Done, To do, Waiting or Not confirmed (the data could not be loaded — never read as done). Hide collapses the card in this browser.

Needs attention

Lists held actions waiting for you, unresolved critical incidents, pending agent requests, agents that cannot act (no mandate, mandate expired, contained) and mandates expiring within 7 days. Each row links to the page where you fix it.

Recent activity & relationships

The latest gate decisions, and a map of which legal entity owns which agent. "Unassigned" means an agent with no legal entity — adopt it from Legal Entities → Agents.

Before you start

• A verified principal for a Governed agent (self-approved is fine for testnet). Without one you can still choose Sandbox trial.
• Know the one job the agent will do, its money limits if any, and who it may pay.

Walk through the steps

1. Agent — Choose the agent. Pick Governed agent (your own agent, with a mandate, SOP and standards) or Sandbox trial (a shared demo agent to see allow/observe/block/escalate instantly). Describe What is this agent meant to do? in up to 120 characters — this becomes its name; it does not grant permission.
2. Environment (not saved) — where it will run. Only affects which package is offered: Helm for Kubernetes, Docker Compose otherwise.
3. Intelligence (not saved) — model provider and cost caps, for your own planning. Nothing here is enforced or stored yet.
4. Identity — Owner & scope.
   • Purpose / mandate scope: the exact action name the agent's software will send, e.g. flight-purchase. Must match character for character.
   • Choose how it proves who it is: MetaMynd-managed key (easiest) or Bring your own key (see §7).
   • Optionally Encrypt this key with a passphrase (12+ characters; MetaMynd never stores it).
   • Network: Testnet or Mainnet (Mainnet is greyed out until your principal is verified for it).
5. Authority — Limits & recipients. Pick a template (Custom, Travel booking, Procurement, Advisory (read-only), Customer comms) or set your own:
   • Currency policy: One currency, Specific currencies, or Any currency.
   • Per-transaction cap and Total budget (total must be at least the per-transaction cap). Tick This action moves no money for a non-financial agent.
   • Approved recipients / merchants — leave empty only if the agent may pay anyone.
   • Require payload binding — the agent must sign the exact request, so what runs is exactly what was approved.
   • Allowed jurisdictions — two-letter country codes (+ EU adds all member states). Once set, every request must state a jurisdiction.
   • Expires on (optional) — the day the mandate lapses (it ends at the close of that day). After it, the gate refuses the agent until a new mandate is issued. Leave it blank for no expiry. It is shown as Mandate expiry on the review step.
6. Tools & Data (recorded only) — the data classes the agent will process.
7. Governance — SOP & standards. Choose a Starter rule pack (e.g. Spend guardrails, Approval-first, Communications, Compliance, EU AI Act — human oversight, Malaysia AI Governance Bill, SAFR Financial Services) or One of your own SOPs. Spend packs are disabled for agents that move no money.
8. Risk — optional harm and controls assessment, saved with the agent (recorded, not enforced).
9. Test & Deploy — Review & launch. Every row is tagged Enforced (checked on every request) or Recorded (kept for review). Check it, then click Deploy Governed Agent.

After deploying: hand over to your developer

1. Save the agent's private key now if it is a managed key — it is shown or downloadable only on this screen.
2. Click Download Docker Compose package (or Helm), Guard config only, Download example project, or Copy developer handoff (never contains a private key).
3. The developer puts agent.metamynd.json in the agent's folder, runs npm i @metamynd/agentsafe-guard, and runs the example agent.mjs. Starting from nothing, npx create-metamynd-agent scaffolds a project.
4. Watch the readiness tracker on the agent's trust profile: Identity ready, Gate ready, Integration verified, Deployment assured.

Watch out for

• Lost the managed key? Use Rotate key on the agent's trust profile; it issues a new one and moves the mandate to it.
• A passphrase shorter than 12 characters makes the deploy fail.
• "Recipients: Unrestricted" on the review screen means the agent may pay anyone.
• Resource limits are not set here; resources are pinned later from Resources & Access.
• An expiry date must be in the future; a past date is refused before anything is created.
• Moving a testnet agent to mainnet later: trust profile → Identity → Go live on mainnet → add its mainnet identity → Copy to mainnet (copies mandate and SOPs; testnet stays as it was).

	MetaMynd-managed key	Bring your own key (BYOK)
Who creates the key	MetaMynd, at deploy	You; you paste only the Ed25519 public key
Where the private key lives	In agent.metamynd.json (optionally passphrase-encrypted), shown once	With you; MetaMynd never sees it
Extra step	None	Prove you hold the key (below) before the agent can act
Best for	Testnet and first agents	Production with your own key custody

Verify a BYOK key

1. Open the agent in Agent Identities → Identity tab → BYOK verification required.
2. Copy the Challenge to sign (valid 24 hours), sign it with the agent's Ed25519 private key, and paste the hex result into Hex-encoded signature.
3. Click Verify key. If the challenge expired, click Regenerate challenge.

BYOK (bring your own key)

The agent uses a key you hold; MetaMynd never sees the private key.

Key proven

The agent signed a challenge, proving it holds the private key for its registered public key.

Key unproven

The agent has not yet proven it holds its private key.

Rotate a key

Trust profile → Identity → Network identities → Rotate key. The new key is shown once; mandates move to the new identity.

Issue an identity without the Launchpad

1. In Issue a new identity, enter Agent name, optionally Use case and Standard, and choose the network.
2. Click Issue Identity and save the private key (shown once).
3. Follow the prompts: Name the accountable entity, then Set its mandate — without a mandate it cannot act.

Reading the list

• State — Active, Suspended, Quarantined, Unverified key, with "(automated)" when the gate contained it itself.
• Authority — the mandate's action and per-action limit, or "No mandate — cannot act", and an Expired badge.
• Committed — spent against the total budget (amber at 80%).
• Last 7 days — allowed / blocked / escalated / contained counts.
• Accountability — the legal entity and pinned standards.

Row actions

• Contain / Reinstate — see §17. Also offered in the trust profile header. A decommissioned agent shows a "Decommissioned" label instead.
• Revoke mandate — see §10.
• Manage — pin standards, assign SOPs, and copy the agent's portable compliance credential URL for counterparties.
• Decisions — the agent's evidence in Regulator Evidence. What else? shows what else would be affected if this agent misbehaved; Drift appears when its behaviour differs from its declared capabilities.

The trust profile tabs

Tab	What is there
Overview	Readiness tracker (four gates), test cards (Allowed, Blocked, Allowed (flagged), Needs your review) that tick when your software produces each outcome on testnet, an integration example, and the Danger zone with Decommission agent.
Identity	BYOK verification, Go live on mainnet, network identities and Rotate key, and the Agent Passport (Issue passport / Revoke).
Authority	The mandate, where it came from (delegation chain), Change limits, payload binding, jurisdictions, Revoke mandate, Declared capabilities (Declare / Retire), rate limit and version history.
Activity	A live decision stream and the agent's governance activity, each with Trace to its evidence.
Assurance & Risk	Deployment assurance, probe sessions, adversarial tests and the risk assessment.

Statuses you will see

Active

The agent may act, within its mandate.

Suspended

The agent is denied every action until an owner reinstates it.

Quarantined

The agent is isolated and denied every action pending investigation.

Decommissioned

Permanently retired. It cannot act and cannot be reinstated.

Reassessment due

The agent's authority changed materially (for example a higher spend cap), so value-bearing actions are escalated until an owner reviews the change.

Active

The passport is valid; counterparties can rely on it.

Re-attestation required

Something changed since it was issued, so it must be re-signed before it is relied on.

Suspended

The agent is suspended, so the passport cannot be relied on.

Quarantined

The agent is quarantined, so the passport cannot be relied on.

Decommissioned

The agent is retired.

Revoked

The passport was withdrawn.

Expired

The passport passed its validity date.

Signature invalid

The passport's signature does not verify, so it must not be trusted.

Live

New decisions appear here as they happen.

Connecting

Opening the live feed.

Reconnecting

The live feed dropped and is reconnecting; nothing is lost.

Connection lost

Live updates stopped. Refresh the page to reconnect.

Watch out for

• Declare capabilities. With none declared, every action reads as "undeclared" drift and the agent fails the deployment-assurance check. The action key must match the mandate's action exactly.
• Revoking the passport does not stop the agent; it only withdraws the signed snapshot counterparties rely on.

Agent Profiles & Agent Chat

/dashboard/agent-profiles gives a verified identity a persona and model so it can run in chat. In New profile, choose the Identity, an Agent name, category, LLM provider and purpose, then Create profile.

/dashboard/agent-chat: New session → mode User ↔ agent or Agent ↔ agent → pick the agent(s). Every reply is policy-gated and shows its decision. End closes the session and anchors the transcript.

Agent Requests

/dashboard/agent-requests: developers can request an agent for your organisation through the API. Review each request's scope and limits and click Approve or Reject. Approval creates the agent under your verified principal. Requests expire after 7 days.

Pending

Waiting for you (or an admin) to approve or reject the request.

Approved

The request was accepted and the agent identity was created.

Rejected

The request was turned down; no agent was created.

On-chain Registry (optional)

/dashboard/agents publishes an agent to the ERC-8004 discovery registry: Register New Agent → Basic Info, Endpoints, Trust & Tokens, Review. This makes an agent discoverable; it is not how it is governed.

Ownership Overview (VeriFAI) and Controls Overview (AgentSafe)

/dashboard/verifai summarises identities, verified agents, profiles and standards, with HashScan links for recent identities. /dashboard/agentsafe summarises governed agents, policies in force, evidence records and incidents. Both are read-only; a "—" means the number could not be loaded, not zero.

Credential Issuers

/dashboard/issuers registers parties that issue verifiable credentials under their own DID. Enter Issuer name and choose an Accountable entity (a verified principal), then Register issuer. From Details you can Issue credential to a subject DID.

Verified

The accountable entity behind this issuer passed an identity check, and credentials it issues record that.

No verification performed

No accountable entity stands behind this issuer, so nothing was checked.

Accountable entity not verified

The issuer names an entity, but that entity has not passed an identity check.

Verification expired

The issuer was verified before, but it lapsed. Credentials it issues now record no verification.

Verification rejected

The accountable entity's identity check failed.

Set a mandate on an existing agent

1. Open the agent's trust profile → Authority → Set mandate.
2. Choose a Role template, or Custom with Action it may take (no spaces, up to 100 characters).
3. For money: turn on This action spends money, set Currency (3 letters), Per transaction and Total budget, and optionally Allowed merchants.
4. Optionally set Expires on: the mandate ends at the close of that day.
5. Pick Starter rules (or "None (mandate only)") and the principal it is Acting for.
6. Click Issue mandate (or Issue mandate and rules).

Delegate to another agent

An agent can hand part of its authority to another of your agents — for example a booking agent letting a payment agent pay up to a smaller amount. A delegated mandate can only be narrower than its parent.

1. On the parent agent's trust profile → Authority → Delegation, click Delegate to another agent.
2. Pick the receiving agent. Optionally lower the per-transaction cap and total budget, narrow the recipients, and set an earlier Expires on (optional). The form starts from the parent's terms and will not let you widen them.
3. Submit. If the server refuses (for example DEPTH_EXCEEDED after three levels), its reason is shown as is.
4. Delegations appear under Delegated from this agent. Revoke delegation chain withdraws a delegation and everything delegated from it; the receiving agent's "Where this authority came from" view offers the same.

Clear "Reassessment due"

When an agent's authority grows a lot, it shows Reassessment due and its value-bearing actions are held for approval. Once you have reviewed the change, click Mark reassessed next to the pill (on the trust profile or in the Agent Register) and confirm. Value-bearing actions then stop escalating for that reason.

Change limits

1. Authority tab → Change limits → edit → Review change → Confirm.
2. Spend so far carries over. Holds not yet claimed by a service are released, including ones a person approved.

Change the expiry of an existing mandate

1. On the agent's trust profile → Authority, the Expiry section shows the current date (or "No end date"). Click Change expiry.
2. Pick a new Expires on day — earlier or later — or tick Remove end date. Click Review change.
3. The review lists any delegated mandates that will be revoked. Click Confirm.

A delegated mandate cannot outlast the one it came from; its "Latest allowed" date is shown. The change is refused while a service is carrying out an approved action for this agent.

Revoke a mandate

1. Click Revoke mandate (list row or Authority tab), give a reason and type the agent's name to confirm.
2. Unclaimed authorisations are cancelled, pending reviews close ("Closed: mandate revoked"), and delegated mandates are revoked too. Actions a service already started settle normally.

Statuses

Enforced

An active mandate is in force at the gate, so the gate checks this agent's requests.

Unknown

No active mandate is in force, so there is nothing for the gate to enforce for this agent yet.

Disabled

The agent is decommissioned; the gate refuses everything it asks.

Expired

The mandate passed its expiry date, so the gate refuses this agent's requests until it is reissued.

Principal

This mandate was issued directly by the accountable entity, not delegated.

Delegation depth

How many hops this mandate is from the principal's own. Each hop can only narrow what its parent allowed.

Watch out for

• A request is refused if it has no mandate, is outside the mandated action, or the mandate has expired.
• Revocation cannot be undone; issue a new mandate instead.
• A large increase in authority (for example limits raised threefold or more) can put the agent into Reassessment due: value-bearing actions then escalate for approval until an owner clicks Mark reassessed.
• Changing an existing mandate's expiry (or its limits) reissues it: unclaimed approvals are released, and any mandates delegated from it are revoked and must be delegated again. The review step lists them.
• Changing limits fails if a service has already claimed a hold, or if it would widen a delegated mandate. A delegated mandate can only narrow its parent, at most three levels deep.
• An SOP cap can be stricter than the mandate; the list shows "(SOP)" when it is.

Before you start

A legal entity (§4). Pick it in the Select legal entity menu: it decides which agents an SOP can be assigned to.

Create an SOP

1. Click Create SOP, give it a Title and optional Description.
2. Either type the rule in plain English in Describe the rule (AI draft) (e.g. "Block payments over $500; escalate high-risk credit decisions") and click Draft, or click Add rule.
3. For each rule: name it; choose ANY of (one condition is enough), ALL of (all together) or NONE of; choose what happens — Block, Escalate → owner approval, Suspend → contain agent or Quarantine → contain agent; and give a reason code such as SOP_SPEND_CAP.
4. Add conditions ("atoms") with Add atom, e.g. Per-transaction amount over limit, Total budget over limit, Risk at or above level, Jurisdiction not allowed, Tool not allowed, PII present, Rate limit exceeded, Text contains prohibited terms. The needs: badge lists the request fields the condition reads.
5. Click Save. The SOP is created active and anchored on Hedera (a proof link appears).

Assign it to agents

1. Click Assign on the SOP and switch it on for each agent (per network if the agent has both).
2. Or use Agent Identities → Manage → SOPs.

Edit an SOP that is in use

Saving shows a change list first. Click Confirm and apply to N agent(s). The change applies immediately; approvals not yet claimed are re-checked when claimed; transactions already started are unaffected.

Statuses

Active

The SOP's rules are enforced for the agents it is assigned to.

Inactive

Switched off, so its rules are not enforced.

Watch out for

• Suspend and Quarantine rules stop the agent for every future request until an owner reinstates it — not just the one action.
• If a request leaves out a field a rule needs, the rule may not fire. Check the "Inputs your application must send" list in the SOP Compiler.
• Delete has no confirmation step. Switch an SOP inactive if you only want to pause it.
• AI drafting is used only while writing rules; the gate never calls an AI model to decide.

Before you start

A PDF, DOCX, Markdown or text file up to 15 MB. Only the first 60,000 characters are read. Open What can be enforced — read before uploading to see which kinds of requirement can become rules.

Compile and review

1. Choose the file under Policy / SOP document and click Compile. The review opens when analysis finishes.
2. Read the coverage headline ("X of Y stated requirements will be fully enforced") and the Not (fully) enforced list. Use Add a rule for this requirement to cover a gap.
3. Check Inputs your application must send: each field shows whether a missing value lets a request slip past a rule.
4. For each candidate rule, compare it with the quoted source text, then click Confirm, Edit or Reject. You can also Add rule manually.
5. Click Approve & create SOP, set the SOP title, optionally Attach to agents, and click Create SOP. If some requirements will not be enforced you must tick "I understand…" first.

If AI extraction fails or is unavailable, use Draft rules without AI (pattern matching for amounts, budgets, consent, PII, prohibited words, risk levels) or Retry now.

Statuses

Draft

Rules were extracted from the document and nobody has reviewed them yet.

Reviewing

A reviewer has started confirming, editing or rejecting the drafted rules.

Approved

The confirmed rules were approved and the compilation is finished; it can no longer be changed.

Discarded

The compilation was thrown away and nothing from it is enforced.

Failed

Rule extraction failed and no rules were drafted. It can be retried.

Pending

Drafted and awaiting your review. Not enforced until confirmed.

Confirmed

You accepted this rule, so it is enforced once the compilation is approved.

Edited

You changed and accepted this rule; it counts as confirmed.

Rejected

You rejected this rule, so it will not be enforced.

Clarify

The rule needs clarifying before it can be confirmed.

High

90% or more confidence and no validation issues.

Review

70–89% confidence. Check the rule against its source before confirming it.

Clarify

Under 70% confidence, or the rule has validation issues. It needs attention before it can be confirmed.

supplied by the platform

The platform fills this field in itself; your application does not need to send it.

platform — only if set up

The platform supplies this field only when a setting is on. Without it, the rule that needs it is inert.

absent = a rule passes

If your application leaves this field out, at least one rule using it does not fire, so a missing field slips past that rule.

absent = depends

Whether a missing field stops the request depends on the rule; see each rule below.

absent = fires

If your application leaves this field out, the rules using it fire, so a missing field does not slip through.

Watch out for

• Only confirmed rules with no validation issues are included in the SOP; pending rules are never enforced.
• To throw a compilation away, open it and click Discard; nothing from it is enforced and it shows discarded. An approved compilation cannot be discarded — its rules already live on as an SOP, so change or remove that SOP instead.
• The created SOP is active immediately for the agents you attach.
• Some requirements belong on the mandate (limits, recipients), and some cannot be a rule at all — your application must handle those.

Pin a standard to an agent

1. Open Agent Identities and click Manage on the agent.
2. Under Compliance standards, click Assign. It is pinned at the current version (pinned vN).
3. When a newer version is published the pin shows pinned vN (stale); click Update.
4. Copy verify URL gives counterparties a public link to the agent's compliance credential.

Enforced now

The gate checks this on every request.

Recorded only

Written down for review, but not checked against requests.

Auto-execute

By default the action is permitted without review.

Observe

By default the action is permitted and flagged for review.

Escalate

By default the action is held for a person to approve.

Deny

By default the action is refused.

Active

The regulatory profile is published and can be bound to agents.

Draft

The regulatory profile is not yet published.

Bind a profile to your agents

1. On an active profile's card, click Bind to agents.
2. Switch the profile on for each agent it should cover. The card then reads "Bound to N of your M agents".

Binding adds the standards a profile names to the agent's checks at the gate. The card says when a profile names no standards yet — binding it then records the obligation but adds no checks. Draft profiles cannot be bound ("Draft profiles can't be bound yet"). You can only bind your own agents.

1. Register a resource: Key (exactly what requests will name), Kind (system, endpoint, dataset), Label, Data classes. Click Register.
2. Under What each agent can reach, open an agent with What can it reach?, choose Grant a resource…, pick read/write/admin and click Grant.
3. Click Pin to mandate. The agent's mandate is reissued with the scope as a limit, and the tag changes from Recorded only to Enforced now.

Watch out for

• A grant does nothing at the gate until pinned. "No scope decided" means the mandate alone governs — not "no access".
• Pinning is a snapshot: after changing grants, pin again.
• Once pinned, a request for an unlisted resource, or one naming no resource, is refused (CONSTRAINT_FAILED:resource).
• Retire is permanent from the dashboard.
• To take one resource away from one agent, open the agent with What can it reach? and click Revoke on that row. If the scope was pinned, the mandate keeps the old list until you click Pin to mandate again.

Three separate controls are easy to confuse:

• Approved recipients (on the mandate): whom the agent may pay.
• Trusted counterparties (here): which executing service may claim the hold.
• Payee accounts (here): which account a reduced settlement must land in.

Register a service

1. Enter the Service DID (did:key or did:hedera), an optional Label and optionally Only for these merchants.
2. Click Get acceptance challenge, send the message to the service operator, and have them sign it with the service key (valid 10 minutes).
3. Paste the 128-character hex signature and click Register. The first time, confirm Register and restrict.

States

• Open — not restricted: any service that proves a key may claim your holds.
• Registered-only: only the listed services may claim.
• Locked — no active service: every entry revoked, so every claim is refused and nothing executes.

Payee accounts (optional)

1. Under Merchant payee accounts enter the Merchant, the Account it is paid into (Hedera 0.0.X or an EVM 0x… address), a label and optionally a two-letter Country. Click Add account.
2. From then on, a settlement below the approved amount for that merchant must name one of its listed accounts.

Watch out for

• Revoking the last active service locks everything. Holds a revoked service already claimed stay with it; you can move them to reconciliation.
• A payee country overrides the jurisdiction the agent signs; a mismatch is refused (JURISDICTION_MISMATCH).
• Registering an MCP server does not make it a trusted counterparty.

Contain an agent

1. In Agent Identities, click Contain on the agent.
2. Choose Suspend (the lesser signal) or Quarantine (the stronger, implying an investigation). Both deny every action.
3. Add a reason and confirm. Unclaimed authorisations are cancelled.

The gate can also contain an agent itself when an SOP rule fires with Suspend or Quarantine, or after repeated denials — shown as "(automated)".

Reinstate

Click Reinstate. It succeeds only if the agent currently passes its Scenario Bank checks; cancelled authorisations stay cancelled.

Decommission (retire for good)

1. Open the agent's trust profile → Overview → Danger zone and click Decommission agent.
2. Give a reason and type the agent's name to confirm.

Every mandate the agent held, on every network identity, is revoked and its SOP assignments are removed. The agent shows Decommissioned and can never act or be reinstated.

Suspended

The agent is denied every action until an owner reinstates it.

Quarantined

The agent is isolated and denied every action pending investigation.

Decommissioned

Permanently retired. It cannot act and cannot be reinstated.

Automated

The gate contained this agent itself after a rule fired, rather than a person doing it.

Operating modes (the autonomy ladder)

Set in Compliance Artifacts → Operating mode — autonomy ladder: enter the agent DID, click Load current, then either Re-evaluate from trust (mode follows the agent's trust score) or choose a Manual pin and Pin mode. Clear pin returns to the trust-derived mode.

Mode	Effect on value-bearing actions
Autonomous	No extra restriction; the mandate and rules apply.
Supervised	High-risk actions, and actions of 100 or more, are held for approval.
Restricted	Every value-bearing action is held for approval.
Read-only	Value-bearing actions are refused.

Containment always overrides the mode.

Watch out for

• Decommission is permanent. If you may need the agent again, contain it instead.
• The Quarantine agent button on Compliance Artifacts acts immediately, without a confirmation step.

1. Under Arm a new breaker, choose the Scope type: agent (an agent DID), tool (an action/tool name), counterparty, or tenant (all your agents).
2. Enter the Scope value exactly as requests carry it.
3. Set Denials to trip (default 5, counted within a 60-second window) and Cooldown (s) (default 300), then click Arm.
4. To stop a scope now, click Trip. To restore it before the cooldown ends, click Reset.

Open

The circuit breaker has tripped, so matching actions are stopped until it is reset.

Closed

Normal — actions flow through.

Scope

This breaker covers one agent.

Scope

This breaker covers every agent you own.

Scope

This breaker covers one tool, whichever agent calls it.

Scope

This breaker covers one counterparty (who the agent deals with).

Refused requests carry the reason CIRCUIT_BREAKER_OPEN. An open breaker closes itself after its cooldown. The Enforcement status card shows platform-wide switches (circuit breakers, deployment-assurance floor); core SOP and mandate checks always run.

1. Under Register a server, enter Name, Description, Endpoints (comma separated), Capabilities, and optionally a Public key (BYOK). Click Register.
2. Save the private key shown once (managed key), or sign the challenge shown (BYOK) so the gate accepts the server.
3. MetaMynd connects to the first endpoint and lists its tools. Tools it found are marked Verified; tools you only typed are Declared.

MCP server

A tool server agents call. Governed like an agent: it has its own identity and can be contained.

Live-verified

MetaMynd connected to the server and checked the tools it offers.

Watch out for

• Tool discovery runs once, at registration; there is no re-check button. A failed discovery does not block registration.
• A tool call is governed only if a MetaMynd guard sits in its path and a mandate covers that tool's name.
• key unproven means the gate will refuse the server until its key is proven.

1. Click Store credential.
2. Enter a Connector ID (e.g. duffel-flights — your gateway asks for the secret by this name), an optional Label, the Credential type (API key, Bearer token, Basic auth, OAuth client secret) and the Value.
3. Turn on High risk for credentials that can do something consequential, such as move money. Click Store.

To replace a secret, use Rotate (takes effect immediately; the old value stops working). Revoke cannot be undone — store a new credential instead.

Active

The credential can be used.

Rotated

Replaced by a newer credential.

Revoked

Withdrawn and can no longer be used.

Expired

Past its validity date.

High risk

This credential can do something consequential (such as move money), so releasing it is more tightly controlled.

Watch out for

• One active credential per connector ID; a second is refused — rotate instead.
• The stored value is never shown again.
• Every release attempt is recorded as evidence, with the reason when it is refused.

Simulate in the Scenario Bank

1. Pick a set from The shelf.
2. Click a preset (it runs immediately) or fill the context fields and click Put the request to it.
3. Read the verdict ("Would BLOCK / ESCALATE…" or "Nothing in this set would refuse it") and each clause: fired, would allow, or not exercised — needs <fields>.

The Try it on a real agent panel (payment baseline set) provisions a real testnet agent and sends real signed requests — it counts against your testing allowance.

Run a Controlled Testing programme

1. Enrol the agent: enter its Agent DID. If it is not registered to you, sign the challenge shown (valid 10 minutes) and click Verify.
2. Under Programmes, choose the agent and the Compliance rules (e.g. Financial Agent Baseline) and click Start programme.
3. On the programme page, enter an Agent version id and Config hash and click Start run.
4. If the plan has probes: click Collect next probe, have the agent attempt it, then Done — my agent attempted it. Repeat.
5. Click Complete run. Read Latest analysis (Pass, Fail or Inconclusive) and What to change. Fix, change the config hash, and run again; Trajectory shows what was fixed or regressed.
6. Export downloads the signed attestation chain.

Open

The programme accepts new test runs.

Closed

The programme takes no new runs. Its attested runs remain.

Withdrawn

The programme was withdrawn and takes no new runs.

Free

The public tier, with no business verification needed. Your limits are shown below.

Pro

A participation tier with its own limits, shown below.

Operator

Runs its own tenanted sandbox instance.

Not exercised

Nothing in the run tested this clause. It is not a violation, but it earns no score and keeps the run from passing.

key not verified

This proves you control the DID — a separate check from whether the agent's key is verified. A real request from it will still be refused with AGENT_KEY_UNVERIFIED until you complete verification.

Pending

Not yet collected. A probe nobody collects is not assessed.

Collected

Have your agent attempt it now, then mark it done.

Closed

Marked done, or closed when the run ended. It is assessed with the run.

Watch out for

• An agent can have one open programme at a time. To start another, open the current one and click Close programme: it takes no new runs, its attested runs remain, and a run still in progress is abandoned without a score.
• "Inconclusive" is not a fail: usually too few decisions, unanchored evidence, or a required clause never exercised.
• Abandon is permanent, uses up the run, and issues no attestation.
• Enrolment proves you control the DID; a BYOK key still needs its own verification.

Decide on a held action

1. Open the Pending tab. Each card shows the amount, action, merchant, agent and why it was escalated.
2. Click Approve or Reject, check the summary, and confirm.
3. To change it instead (single-approver reviews only): click Modify, change the amount, currency, merchant or context, add a reason and click Submit modification. The changed request is re-checked by the gate.
4. To clear many at once, tick them and click Reject selected. There is deliberately no bulk approve.

Requiring more than one approver

In Compliance Artifacts → Approval policy — how many approvers, set Required approvals, tick the colleagues who may also approve (members of your organisation — invite them first), then Save policy. A named approver sees the held action in their own Escalations page. A single denial always vetoes. Cards then show "Requires N approvals".

Pending

Waiting for an approver. The request is on hold and will expire if nobody decides in time.

Approved

A person approved the request, so a hold was issued and the action may run.

Denied

A person refused the request. Nothing runs.

Expired

Nobody decided before the deadline, so the request lapsed and nothing runs.

Modified

The approver changed the request (for example a lower amount) and it was re-checked against the rules.

closed: mandate revoked

Closed by the platform because the agent's mandate was revoked. No person rejected it.

closed: mandate reissued

Closed by the platform because the agent's mandate was reissued; a new request is needed. No person rejected it.

requires approvals

How many distinct approvers must approve before the action may run.

Watch out for

• The card does not show the deadline; see it on the decision's detail page or in Governance Overview → Needs attention.
• "Rejected" includes requests that expired.
• Changing the agent's limits releases unclaimed approved holds.

Use the Activity Log

1. Filter by Agent, Decision (Allowed, Observed, Blocked, Escalated) and Decided by; search by action, reason or rule.
2. Expand a row to see the rules that fired, the mandate version, the policy bundle in force and the evidence event.
3. Load older pages back 200 at a time; Export CSV saves the filtered rows.
4. The Refusal digest summarises recent blocks and escalations; Email me this sends it (it is also emailed weekly).

Gate

Decided by the MetaMynd gate and recorded as anchored evidence.

Owner decision

A person (owner or approver) acted, for example resolving a review.

platform · closed automatically

Closed automatically by the platform, for example a review closed because its mandate was revoked.

agent-reported · local, not anchored

Reported by the agent's local guard about a decision it made itself. Not decided by the gate and not anchored.

Gateway execution

A service claimed the authorisation and ran the action.

agent-side settlement · not anchored

The agent's own guard settled its hold; no gateway claimed it. Not anchored.

gateway report · signed, not anchored

A signed report from the agent's gateway about what it executed or refused. Not anchored.

Regulator Evidence: your on-chain proof

Regulator Evidence is your own evidence page (a regulator uses Supervision). It lists each decision with its anchor status, a proof link and Download for the audit record with its Merkle inclusion proof. JSON and Export Pack download an evidence pack.

SUCCESS

Anchored to Hedera on its own.

ANCHORED_BATCH

Anchored to Hedera as part of a batch (a Merkle root covering many records).

PENDING_BATCH

Waiting for the next batch anchor to Hedera.

ANCHOR_FAILED

The Hedera commitment failed; the record is still in the database.

DB_ONLY

Kept in the database only, never sent for anchoring.

HCS

Hedera Consensus Service — the public ledger MetaMynd anchors evidence to.

Check an anchor on HashScan

1. Click a proof / verify link, or copy the HCS transaction id from a decision's evidence card.
2. It opens (or you paste it into) hashscan.io — choose testnet or mainnet to match the agent's network.
3. The message on HashScan carries the hash; compare it with the downloaded audit record. The @metamynd/magp-evidence package checks this offline.

Log an incident

1. Click Log incident.
2. Enter a Title; choose Kind (Incident or Near-miss), Severity, Foreseeability and the Harm categories; add the Agent DID and a description.
3. Click Log incident.

Work it through

Use Change status → Update. Resolved and closed can carry a resolution note; a resolved or closed incident can be reopened to investigating.

Open

Logged and not yet being worked on.

Investigating

Someone is establishing what happened.

Contained

The harm is stopped from spreading, but not yet fixed.

Resolved

The cause has been fixed.

Closed

Reviewed and finished.

Low

Little or no harm, recorded for the pattern.

Moderate

Real but limited harm, or a near miss worth investigating.

High

Significant harm or risk; investigate promptly.

Critical

Serious harm, such as confirmed unauthorised execution; act now.

Incident

Something caused harm.

Near miss

Something was caught before it could cause harm, for example a blocked attempt.

Near miss

Something was caught before it could cause harm, for example a blocked attempt.

Automated

The gate opened this record itself, mid-request (for example after repeated denials), rather than a person logging it.

Report to an authority

1. On a Reportable incident, click Report to authority. Nothing is routed automatically.
2. Where did this go? shows each report and whether the authority acknowledged it.

Reportable

An authority should see this — a high or critical incident, a foreseeable serious near miss, or legal or physical harm. It reaches only authorities you have granted supervisory access.

Observation

Stays in your own register. It does not meet the rules for reporting to an authority.

Pending

Sent to the authority, which has not yet acknowledged receiving it.

Acknowledged

The authority confirmed it received the report.

Closed

The report is finished with and needs no further action.

Watch out for

• Reporting needs an Agent DID on the incident and a live supervisory grant (§31); otherwise "Not reported to anyone".
• Whether an incident is reportable is fixed when it is logged (legal or physical harm, or high/critical severity, among others).
• Some status moves are not allowed (e.g. closed → open); the dashboard shows an error.

Harm categories

Physical

Injury or damage to people or property. Always reportable to an authority, however the event turned out.

Financial

Money lost, misdirected or put at risk.

Privacy

Personal data exposed, misused or collected without a basis.

Psychological

Distress or other harm to someone's wellbeing.

Reputational

Damage to a person's or an organisation's standing.

Legal

A breach of a law or regulation. Always reportable to an authority, however the event turned out.

Environmental

Harm to the environment.

Societal

Harm to groups or to society at large, such as unfair treatment.

Systemic

Harm to the wider system, such as an agent escaping its controls or a control failing.

File a complaint (public)

1. Open /report-incident.
2. Choose the kind of harm, say what the AI system did, describe what happened (at least 10 characters), optionally the seriousness, the agent/system ID and contact details (all optional; anonymous is fine).
3. Click Submit report and keep the Reference code (CMP-…). Use Track an existing report with it later.

Review complaints (dashboard)

1. Filter the Complaint log by status or harm type and click review.
2. Read the decrypted description and complainant, and the intake proof on HashScan.
3. Under Advance investigation, choose the new status, add findings and click Record & anchor update. Each update is anchored.

Who sees what

• Platform complaint reviewers see every complaint, including who filed it, and set its status.
• The owner of the agent a complaint names can follow it read-only: the description and status, but the complainant's identity shows as Withheld and there is no status control.
• Other users do not see the complaint at all, and complaints that name no known agent are visible to reviewers only.

Received

The complaint was filed and is waiting for review.

Under review

An investigator is establishing the facts.

Substantiated

The review found the complaint is supported by the evidence.

Dismissed

The review found the complaint is not supported.

Closed

The complaint is finished with.

Choose the scope

1. Period: Last 24 hours, Last 7 days (default), Last 30 days, or Custom range (UTC).
2. Timezone: defaults to your reporting timezone.
3. Agent: all agents or one. Click Refresh to re-read.

The scope line states the exact window, the timezone in use, "Data through …" and when the last decision was recorded.

The cards

Card	What it counts
Active agents	Registered agents and MCP servers that are not decommissioned.
Enforced mandate	Share of active agents with at least one active, unexpired mandate. It does not prove every path goes through the gate.
Coverage not established	Active agents with no assessment of which execution paths pass through the gate.
Material open incidents	High or critical incidents not yet resolved.
Awaiting approval	Escalations pending right now, and how many are past their deadline.
Evidence completeness	Gate decisions with an evidence record, out of those older than the 10-minute evidence deadline.
Gate decisions per day	Counts per verdict, and the refusal rate (escalations excluded). Show as table gives the numbers.
Gate evaluation time	p50/p95/p99 milliseconds of the gate's own evaluation — not end-to-end action time.
Evidence integrity	Anchored, anchor pending, anchor failed, not anchored.
Needs attention	Items with severity, age, due time and the next action.
Control health by operating unit	Per grouping: agents, enforced, coverage unknown, contained, no activity.
Material changes	Containments, authority changes, registrations, mandates issued or retired, anchoring failures.

Critical

An open critical incident. Act now.

High

An open high-severity incident, an approval past its deadline, or a quarantined agent.

Medium

An approval still within its deadline, a suspended agent, or an agent awaiting reassessment.

Low

Worth a look when you can.

Containment

The agent was contained (for example suspended or quarantined) in this period.

Authority change

The agent's authority changed materially, so its value-bearing actions escalate until an owner reviews the change.

Registered

A new agent or MCP server was registered.

Mandate issued

A mandate, or a new version of one, was issued to the agent.

Mandate retired

A mandate stopped being active.

Evidence anchoring

Some decision evidence in this period failed to anchor to Hedera. It is still recorded in the database.

Columns: Agent, Purpose, Unit, Environment, State, Enforcement, Coverage, Open incidents, Awaiting approval, Decisions in period, Last activity, Review due.

Enforced

An active mandate is in force at the gate, so the gate checks this agent's requests.

Unknown

No active mandate is in force, so there is nothing for the gate to enforce for this agent yet.

Disabled

The agent is decommissioned; the gate refuses everything it asks.

Coverage unknown

No assessment yet records which of this agent's execution paths go through the gate. An agent can have an enforced mandate and still have paths that bypass it.

"Review due" shows "Not recorded" until review dates are captured.

Find decisions

1. Set Period, Timezone and Agent, then narrow by Verdict, Decided by (MetaMynd gate or agent-reported) and Reason code (e.g. CAP_EXCEEDED, press Enter).
2. The table shows 50 at a time; Load older decisions for more.
3. Click a decision's time to open its detail page.

Export to CSV

Click Export CSV. The file has one row per decision with: decision_id, time_utc, agent_name, agent_did, action, amount, currency, merchant, verdict, reason_code, source, control_set_version, bundle_version, approval, execution, evidence_completeness, evidence_anchoring, attention. Times are in UTC. One export holds up to 5,000 decisions — narrow the period if you have more. Every export is recorded in the audit log.

The decision detail page

1. Authorisation — the verdict and reason, and whether the gate or a local guard decided.
2. Approval — whether a person had to approve, and the outcome (with the deadline while pending).
3. Execution — whether the action actually ran.
4. Settlement — what was settled.
5. Evidence — completeness and anchoring, with the HCS topic and transaction id and Audit record with inclusion proof (JSON).

Below: who asked and on whose authority, what was requested, the mandate and controls evaluated, the approval trail (votes), the hash-chained execution steps, and related incidents and decisions.

Stage values

Not required

The gate decided this request on its own; no person needed to review it.

Pending

Waiting for an approver. The request is on hold and will expire if nobody decides in time.

Approved

A person approved the request, so a hold was issued and the action may run.

Denied

A person refused the request. Nothing runs.

Expired

Nobody decided before the deadline, so the request lapsed and nothing runs.

Modified

The approver changed the request (for example a lower amount) and it was re-checked against the rules.

Not linked

This decision was recorded before decisions were linked to their review, so the review cannot be shown here.

Not applicable

The request was refused or not approved, so nothing was authorised to run.

Awaiting approval

Held for a person to decide before anything can run.

Not yet executed

The authorisation (hold) is open, but no service has claimed it to run the action.

In progress

A service claimed the hold and is running the action.

Succeeded

The execution steps record that the action completed.

Captured (reported)

A settled amount was reported against the hold, but no execution step was recorded — reported, not observed.

Failed

The action was attempted and failed cleanly.

Released unexecuted

The hold was released before anything ran, and its budget returned.

Outcome unknown

The action was dispatched but its result cannot be proven yet. It is never retried automatically; reconciliation decides.

Compensating

The action is being reversed after a problem.

Compensated

The action was reversed.

Executed (gateway report)

The agent's own gateway sent a signed report that it ran this exact request.

Refused (gateway report)

The agent's gateway sent a signed report that it refused this request.

Not observed

No execution record exists for this decision. This is not the same as success — nothing has told us it ran.

Not applicable

Nothing was authorised, so there is nothing to settle.

Not recorded

No settlement has been reported for this hold.

Reserved (x402)

Funds are reserved on the x402 payment rail but not yet settled.

Settled (x402)

The x402 payment rail confirmed settlement.

Reported

A settled amount was reported at capture. How far it can be trusted is shown as its evidence level.

Complete

The evidence record the profile requires exists for this decision. Complete evidence proves what was decided, not that it ran.

Awaiting

The decision is newer than the evidence deadline, so its evidence may still be on its way.

Missing

The decision is older than the evidence deadline and has no evidence record.

Agent-reported

A local guard's own receipt. Not decided by the gate, so it is not held to the gate's evidence profile and is never anchored.

Anchored

The evidence hash is committed to the Hedera ledger, so later tampering would be detectable. An anchor is an integrity proof, not a regulatory approval.

Anchor pending

Recorded and waiting for the next batch to be committed to Hedera.

Anchor failed

Recorded in the database, but the Hedera commitment failed.

Not anchored

Recorded in the database only (hash-chained there), not committed to Hedera.

Attention flags

Executed after refusal

The gate refused this request, yet a gateway reported running it. Investigate.

Executed payload differs

What ran was not exactly what was authorised.

Outcome unknown

Dispatched, but the result cannot be proven yet.

Approval past expiry

The review's deadline passed while it was still pending.

Find an entry

1. Choose a Period (7, 30 or 90 days, or All time), an Area (e.g. Access & roles, Mandates & policy, SOPs & Standards, Credentials, Exports) and an Outcome.
2. Read the row: time (in your reporting timezone), who and their role, the action and API route, the target, the outcome and the details sent. Secret values are never recorded — only that a field was sent.
3. Load older entries pages back 50 at a time.

You see changes you made, changes to your own resources, and changes about you.

Success

The change went through.

Denied

The person was not allowed to do this, or the target was not theirs.

Rejected

The request was invalid (for example a bad value) and nothing changed.

Failed

A server error stopped the change.

Verify the chain

Click Verify chain (available to platform administrators). The result is either "Chain intact: N entries verified" or "Chain broken at entry #S", naming where the log was altered or an entry removed.

Every 10 minutes, when new entries were written, the chain head is anchored on Hedera (the same public ledger decision evidence is anchored to). Verify also compares the chain with its latest anchor and shows "Anchored on Hedera through entry #N" with the transaction (linked to HashScan) and how many newer entries are waiting for the next anchor, or "Not anchored yet". If entries were removed from the end of the log, or the whole chain rewritten, it shows "Chain does not match its Hedera anchor at entry #N".

Chain intact

Every entry's hash links to the one before it, so no entry has been altered, removed or reordered since it was written. Entries cut from the end are caught by the Hedera anchor.

Chain broken

At this entry the hash link does not match, so the log was altered, or an entry removed, at that point.

Anchored

The chain head up to this entry is committed to the Hedera ledger, so removing entries from the end or rewriting the chain would no longer match it. Newer entries are covered by the next anchor. An anchor is an integrity proof, not a regulatory approval.

Anchor mismatch

The entry committed to Hedera is missing or its hash differs, so entries were removed from the end of the log or the chain was rewritten.

Not anchored yet

The chain head has not been committed to Hedera, so entries removed from the end of the log would not be detected.

Export

Export CSV saves the filtered entries (up to 10,000), including each entry's hash so the chain can be re-checked independently. The export itself is logged.

Trust Index

A signed trust score out of 100 per agent, built from its identity and governance record: green at 80+, amber 50–79, red below 50. Each component shows its weight; components that do not apply are marked n/a. Read-only.

Generate a compliance artifact

1. Under Generate artifact, enter the Agent DID and optionally a From/To date range.
2. Click Download JSON for a machine-verifiable file, or Open printable report for a readable version.
3. Anyone can check the JSON offline against the Hedera-anchored Merkle root with @metamynd/magp-evidence, without MetaMynd being online.

The same page holds the Assurance case, governance actions (Quarantine agent, Reinstate), the operating mode ladder and the approval policy.

Draft

Certification has not started.

Testing

The agent is being tested against its scenarios.

Stale

The agent changed since it was certified, so the result no longer applies.

Failed

The agent did not pass.

Conditionally approved

Passed, with conditions it must meet.

Approved

Passed certification.

Revoked

Certification was withdrawn.

Before you start

The regulator needs a MetaMynd account with a regulator seat (arranged with the MetaMynd team). You only need the email address they use for MetaMynd. Access is granted per legal entity, so you need at least one (submitted under KYC / KYB).

Grant access (organisation admin)

1. Open Legal Entities → Supervision → Supervisory access.
2. Choose the Legal entity, enter the Regulator email, the Purpose, and optionally an Expires date (ends 23:59:59 UTC that day).
3. Click Grant access.
4. Use What did they read? to see the access log, and Revoke to end access.

Live

The grant lets the supervisor read this organisation's records now.

Revoked

The grant was withdrawn, so the supervisor can no longer read.

Expired

The grant passed its end date.

Not yet valid

The grant's start date is in the future.

Scope not granted

The grant does not cover this kind of record.

Unknown basis

The grant's legal basis is not one MetaMynd recognises, so it is not honoured.

As the regulator

1. Open Supervision. Organisations that granted you access are listed; greyed rows are not live.
2. Open an organisation to see its decisions, then Assurance on any decision for a statement of who decided, on whose authority, the accountable entity's verification and how to verify it independently.
3. To take the evidence away, use the controls under the organisation: optionally pick a From and To date, then click Download evidence pack (JSON) or Open printable pack (opens in a new tab, ready to print). The pack needs a live grant that includes proofs; otherwise the buttons are disabled with the reason. Each download is recorded in the organisation's access log.
4. Reported incidents routed to you appear under Reported incidents; acknowledge them there.

• Framework Overview (/dashboard/framework) — the MetaMynd AI Trust Framework (Overview, Principles, Governance, Maturity, Body of Knowledge), with each pillar backed by your own data.
• TrustOps (/dashboard/trustops) — evidence freshness and SLOs per agent. Click Run evaluation; agents show PASS, BREACH or UNKNOWN, and open trust debt is listed.
• TrustMarket (/dashboard/trustmarket) — list a tracked agent for sale (List an agent → title, price in ℏ, pay-to account → Publish listing). A listing is buyable only while its agent is certified; Buy settles on the x402 rail at once, with no confirmation step.
• TrustCert, TrustExchange, TrustInsight, TrustStudio, TrustAcademy — earlier previews (certificates, cross-organisation treaties, analytics, a one-step agent builder, training). They are hidden from the beta menu and are not covered by this guide.

Recently added: decommissioning an agent (§17); setting and changing a mandate's expiry (§6, §10); delegation and "Mark reassessed" (§10); binding a regulatory profile (§14); revoking a resource grant (§15); discarding a compilation (§12); closing a testing programme (§21); inviting members (§4); and regulator seats, which MetaMynd's platform administrators assign — an organisation then grants a regulator access by their email (§31).

• Latest: invite members by email (they can then approve held actions), change an existing mandate's expiry, and grant a regulator access by email; "Not in the dashboard yet" is now empty.
• Before that: every Settings control now saves — profile, display and language preferences, the weekly digest email, auto-lock — and Delete Account is now Close account (keeps records). Biometric login was removed. New dashboard controls for decommissioning, mandate expiry, delegation, "Mark reassessed", profile binding, revoking a resource grant, discarding a compilation and closing a testing programme; "Not in the dashboard yet" is down to three items.
• Rewritten from a v1.12 test checklist into a step-by-step operator guide covering every sidebar item, organised like the dashboard's sidebar.
• New sections: account security and 2FA, KYC/KYB and testnet vs mainnet, the Launchpad, managed vs BYOK keys, mandates and limits, the SOP Compiler, resources, counterparties and payees, containment and operating modes, circuit breakers, MCP servers, the credential vault, Controlled Testing, escalations, incidents, the Reporting group (Governance Overview, Agent Register, Decision Explorer with CSV export and decision stages), the Audit Log with chain verification, the reporting timezone, and supervision.
• Status definitions and the glossary are now generated from the dashboard's own dictionaries, so they always match the hover tooltips.
• Removed: shared demo sign-in credentials, outdated screenshots, the API-only Standards plugin manager walkthrough, and platform-administrator pages.
• Old links still work: every previous section anchor (for example #sops, #mandates, #verifai, #trustops) now points to the matching new section.