What is this?
This is a live governance sandbox. You're simulating what happens when an AI agent tries to spend money, call an API, or interact with an external system — and watching a set of rules decide whether it's allowed or blocked. Every outcome generates a cryptographic receipt, like a tamper-proof paper trail.
Left panel — Configure the action
- Action Type — what the agent is trying to do. Transfer sends USDC to a wallet. Swap exchanges tokens. api_call hits an external URL. data_purchase buys a dataset. model_inference runs an LLM.
- Cost (USDC) — how much it costs in dollars. Try $60 to trigger the budget ceiling, or $30 with the approval gate on.
- Recipient / Target — where the money or request goes. Auto-fills when you change the action type.
- Agent ID — the name of the agent taking the action. This is what gets tracked and scored over time.
Left panel — Policy toggles
- Budget ceiling — hard blocks anything over $50. No exceptions.
- Rate limit — blocks if the agent exceeds 10 actions in a minute. Prevents runaway loops.
- Business hours gate — blocks outside 09:00–18:00 UTC. Useful for time-restricted autonomous agents.
- Human approval gate — requires sign-off for anything over $25. Toggle this on, set cost to $30, and hit Execute to see it fire.
- Base anchor — not a blocker. When on, every allowed receipt gets its hash written to the Base blockchain as permanent proof.
Right panel — What comes back
- Receipt — the full cryptographic record: action, cost, allowed/blocked status, SHA-256 output hash, and EIP-712 signature. This is the tamper-proof paper trail.
- Policy Timeline — a step-by-step walkthrough of the 5 governance stages. Shows exactly where a block happened if one fired.
- Policy Checks — a table of every rule, its pass/fail result, the exact numbers that triggered it, and what it means in plain English.
Try a scenario:
Action Config
Policy Rules
Budget ceiling
Blocks any action costing more than $50 USDC — hard limit, no exceptions
Rate limit
Blocks if the agent fires more than 10 actions in a 60-second window
Business hours gate
Blocks actions outside 09:00–18:00 UTC — try enabling this now to see it fire
Human approval gate
Requires human sign-off for any action over $25 USDC — toggle on + set $30 to trigger
Base anchor
Pins each allowed receipt hash onto the Base blockchain as permanent on-chain proof
Governance engine ready
Configure an action on the left — or pick a scenario above — then hit Execute to run the policy checks and generate a cryptographic receipt.
What is this?
This is a running scoreboard of every action executed in this session. As you use the Run Simulator, this page fills up automatically. It answers the question every CTO asks: "How do I know what my agents are actually doing?"
Total Runs
Every time you hit Execute, this increments. Gives a sense of agent activity volume.
Allowed
Actions that passed every policy check and were permitted to execute.
Blocked
Actions stopped by a governance rule. The ratio of blocked to allowed is your risk signal.
USDC Governed
The total dollar value of allowed actions. This is what an enterprise buyer cares about — "$X has passed through our governance layer."
The activity feed
- Each card shows the action type, cost, the agent that ran it, and a short version of the receipt hash so you can trace it back.
- Green tick = allowed through all policy checks. Red X = stopped by governance. The reason is shown in the card.
- Most recent action is always at the top. Hit Clear to reset the session for a fresh demo.
Live Governance Dashboard
Real-time metrics across all agent actions in this session
0
Total Runs
0
Allowed
0
Blocked
0.00
USDC Governed
Activity Feed
No actions yet. Use the Run Simulator tab to see live activity appear here.
What is this?
This is an AI agent's trust score — a number from 0 to 100 that represents how reliably and safely it has behaved over time. Think of it like a credit score, but for autonomous software. The higher the score, the more autonomy the agent is granted.
The ring and tier
- The big number is the agent's current score. It updates live every time you run an action in the simulator.
- The ring fills proportionally — a score of 85 fills 85% of the circle.
- Bronze → restricted to low-value actions only. Silver → standard limits apply. Gold → eligible for autonomous high-value actions. Platinum → all governance gates unlocked.
The four score dimensions
- Success rate (40%) — what percentage of this agent's actions were allowed through. The biggest factor.
- Output quality (30%) — how clean and valid the agent's outputs were.
- Reliability (20%) — consistency over time. Does it behave the same way repeatedly?
- Recency (10%) — how recently the agent has been active. Inactive agents score lower here.
The EIP-712 ScoreCredential JSON
- This is the actual signed credential that would be issued to the agent — formatted as a W3C Verifiable Credential.
- It contains the score, all four dimensions, the issuer (
did:web:attestify.os), a timestamp, and a cryptographic signature. - Any system that receives this credential can verify it without calling Attestify at all — the math is self-contained. This is how agents prove their trustworthiness to new systems they've never interacted with before, the same way a credit score works when you apply for a loan.
Agent Credential Score
EIP-712 ScoreCredential — 4-dimension weighted trust model
Trust Score
--
/ 100
—
Run actions to build your trust score
Score Dimensions
Weights: Success ×0.40 · Quality ×0.30 · Reliability ×0.20 · Recency ×0.10
EIP-712 ScoreCredential
This is the portable signed credential that gets issued to the agent. Any downstream system can verify it without calling Attestify — the signature and score are self-contained.