Agent Arena

Govern your AI agents before they act.

BYOK + AION token reduction + KINE execution governance.

Connect an agent, review the result, and keep the receipt trail visible.

Agent Protection

Live BYOK

Provider relay with KINE governance

Token Reduction

Estimated

AION cuts unnecessary inference work

Receipt Trail

Measured

Runtime events and receipts stay visible

Developers

KINE SDK Alpha and governed Agent runtime

The local TypeScript SDK Alpha exposes versioned inspection and handoff contracts. For the controlled BYOK demo, a provider key is used for one governed request while AION reduces context and KINE records the decision. The key is not intentionally stored in browser storage, application storage, structured records, or activity events. The SDK does not hold wallet keys, sign, broadcast, or provide production remote delivery.

LIVE RELAYLOCAL DEMOREAD-ONLY / NO WALLET ACTIONBLOCKED BEFORE PROVIDER CALLESTIMATEDMEASURED

KINE Decision Summary

Decision

BLOCK

Governors

Identity, Risk, Policy, Budget, Evidence

Required action

Proceed with governed task

Policy findings

EXTERNAL_CALL_BLOCKED_BY_EVIDENCE

Token Economy Summary

Budget status: demo within budget

Estimated tokens avoided: 39

Measured provider tokens: Not available yet

Estimated cost avoided: Not available yet

Gate decision: EXTERNAL_CALL_BLOCKED_BY_EVIDENCE

Structured Record / History Summary

Record ID

Run governed task to generate one

Decision ID

Pending

History entry

Pending

History size

0 records

Next steps

Move from governed runtime to runtime intelligence and evidence.

Sessions Supporting Evidence Feedback

Advanced runtime details

Provider, model, key, task, and budget

Provider selector

Server-side relay path with an ephemeral BYOK key and normalized usage telemetry.

Model selector

KINE server relay → OpenAI Responses API

Ephemeral API-key field

Use a dedicated project-scoped key with spend limits. Refresh clears the field. The key is transmitted to KINE for this request only and is not intentionally persisted, logged, emitted, or included in receipts.

Token budget input

Budget governor input for the inference-necessity gate.

Task input

LIVE BYOK RUNTIME: OpenAI only through the KINE server relay.

LOCAL DEMO: non-live providers and local-deterministic paths.

Inference-necessity gate

Task received

Yes

External inference necessary

Yes

Deterministic answer exists

Reusable context exists

Duplicate context removed

Yes

Irrelevant context removed

Yes

Token budget allows

Yes

Policy allows

Yes

Evidence satisfied

Gate decision

EXTERNAL_CALL_BLOCKED_BY_EVIDENCE

Deterministic steps

Task normalized · Candidate context assembled · Duplicate context removed · Irrelevant context removed · Budget compared · Policy checked · Evidence checked

Reused context refs: None yet

Token reduction evidence

Original context hash

kine_6228280fa7de81c4

Reduced context hash

kine_1f2e074dddafc136

Original chars

415

Reduced chars

258

Baseline estimated input tokens

104

Optimized estimated input tokens

Estimated input tokens avoided

Estimated reduction %

38%

Sections retained

task, provider, model, budget, governance

Sections removed

redundant-task-echo, irrelevant-ui-noise

Provider usage result

Run a governed task to record measured provider usage or a blocked-before-call outcome.

Governor decisions

Selected governors: Risk, Budget, Policy, Identity, Evidence

Kernel decision: ALLOW

Final visible decision: WARN

Risk governor: ALLOW

Budget governor: ALLOW

Policy governor: ALLOW

Identity governor: ALLOW

Evidence governor: ALLOW

Required action: Run governed task to inspect the action.

Receipt and ledger entry

Run a governed task to prepare a receipt and ledger entry.

Recent token ledger

Browser-local records only. Existing demo records remain intact.

0 records

No token ledger records yet. Run the governed task to create the first browser-local entry.

Truth boundary

The API key is transmitted securely to KINE for the active request only.
The key is used in volatile request memory and is not intentionally persisted, logged, stored in receipts, or emitted through runtime events.
Use a dedicated project-scoped key with spend limits and rotate or revoke it after testing.
OpenAI is the only live relay path in this beta and it runs through the KINE server route.
Non-live providers remain contract-only and are blocked before provider execution.
No blockchain transaction, wallet action, or autonomous external tool execution is allowed here.

View runtime events View evidence boundary

LIVE BYOK runtime active. The relay keeps provider execution behind KINE.

KINE Runtime

Shared activity status

One browser-local activity context supports wallet posture, Action Review, Guardian monitoring, and History.

View activity details

Layer

Layer 1 Website Runtime

Intent

Observed session

Governors

Identity, Policy, Risk, Budget, Evidence

Last decision

No decision yet

Receipt

No receipt yet

Decision: No decision yet

Events

Boundary

Read-only beta / no custody / no hidden execution

Governed agent journey

Task received → Context reduced → Governors evaluated → Decision → Structured record

This is the story the user should feel first: a governed task enters the kernel, AION reduces context, KINE evaluates the action, and a structured record captures the result.

Awaiting governed task

Agent Execution Timeline

How AION Guard governs every agent action from proposal to execution

Agent Proposes

Step 1

AI agent requests action (transfer, swap, approval, etc.)

AION Previews

Step 2

Simulate execution path and predict consequences

KINE Classifies

Step 3

Determine risk level: ALLOW / WARN / BLOCK

Policy Governs

Step 4

Apply agent-specific permission rules

User Alerted

Step 5

Notify via mobile, Telegram, Slack, email, webhook

Execution Allowed or Blocked

Step 6

Safe actions execute. Risky actions are blocked.

Receipt Stored

Step 7

Immutable audit record for compliance and transparency

Every Action is Governed

AION Guard ensures that no agent can execute dangerous actions. Keys authorize. Agents propose. KINE governs.

Agent action queue

Proposed actions awaiting governance

Agent Action Queue

Fixture-backed proposed actions with KINE classification

Clawdbot Treasury Agent->Ethereum

Rebalance treasury funds to higher yield protocol

Transfer 5000 USDC to Compound Finance

Amount:5000 USDC($5,000)

Protocol:Compound

ALLOW

KINE REASON CODE

TREASURY_REBALANCE_APPROVED_PROTOCOL

2 minutes ago->RCPT-AG-20260518-001

Allowed

DeFi Yield Agent->Polygon

Approve USDC spending for yield farming

Approve unlimited USDC to 0x9e5C...unknown

Amount:Unlimited

Protocol:Unknown

BLOCK

KINE REASON CODE

UNLIMITED_APPROVAL_UNVERIFIED_SPENDER

POLICY VIOLATIONS

Unlimited allowance to unverified contract
Spender not in allowlist
No security audit available

14 minutes ago->RCPT-AG-20260518-002

Execution Blocked

Trading Strategy Agent->Arbitrum

Execute arbitrage swap

Swap 2 ETH for USDC via Uniswap

Amount:2 ETH($7,200)

Protocol:Uniswap V3

WARN

KINE REASON CODE

HIGH_SLIPPAGE_DETECTED

POLICY VIOLATIONS

Slippage exceeds 1% policy threshold

1 hour ago->RCPT-AG-20260518-003

Invoice Payment Agent->Base

Process vendor invoice payment

Transfer 500 USDC to Acme Corp

Amount:500 USDC($500)

Protocol:Native Transfer

ALLOW

KINE REASON CODE

ALLOWLISTED_VENDOR_PAYMENT

3 hours ago->RCPT-AG-20260518-004

Allowed

Clawdbot Treasury Agent->Ethereum

Bridge funds to Arbitrum for lower fees

Bridge 10,000 USDC via unknown bridge

Amount:10,000 USDC($10,000)

Protocol:FastBridge Protocol

BLOCK

KINE REASON CODE

UNVERIFIED_BRIDGE_PROTOCOL

POLICY VIOLATIONS

Bridge protocol not in approved list
No security audit available
Contract age < 30 days

5 hours ago->RCPT-AG-20260518-005

Execution Blocked

DAO Operations Agent->Ethereum

Execute approved governance proposal

Distribute 15 ETH grant to 5 recipients

Amount:15 ETH($54,000)

Protocol:Multi-sig

WARN

KINE REASON CODE

HIGH_VALUE_REQUIRES_APPROVAL

POLICY VIOLATIONS

Transaction amount exceeds $10,000 auto-approval limit

8 hours ago->RCPT-AG-20260518-006

Default policy rules

What agents can and cannot do

Agent Permission Policy Builder

Define what each agent can and cannot do

Limits

Limit agent to maximum 10 transactions per hour

Policy Enforcement

All policy rules are enforced before agent actions reach execution. Critical violations result in immediate blocks.

Agent identity

Connected agents under oversight

Agent Registry

Connected agents with KINE governance oversight

Filter by status:

Clawdbot Treasury Agent

Treasury Agent

Active

Multi-chain treasury rebalancing and yield optimization

Connected Wallet0x742d...0bEb

ChainEthereum

Trust Score94/100

KINE Protection

Active

Last Action

Treasury rebalance: 1000 USDC → Compound

2 hours ago

127

Proposed

119

Allowed

Warnings

Blocked

DeFi Yield Agent

DeFi Agent

Restricted

Automated yield farming and liquidity provision

Connected Wallet0x8B3E...4a9e

ChainPolygon

Trust Score71/100

KINE Protection

Monitoring

Last Action

Attempted unlimited USDC approval to unknown spender

14 minutes ago

Proposed

Allowed

Warnings

Blocked

Invoice Payment Agent

Payments Agent

Active

Automated vendor payments and expense processing

Connected Wallet0x1f98...F984

ChainBase

Trust Score98/100

KINE Protection

Active

Last Action

Paid invoice #3847: 500 USDC to Acme Corp

1 day ago

243

Proposed

241

Allowed

Warnings

Blocked

Trading Strategy Agent

Trading Agent

Sandboxed

Algorithmic trading and portfolio management

Connected Wallet0xC02a...6Cc2

ChainArbitrum

Trust Score0/100

KINE Protection

Monitoring

Last Action

Sandbox testing: 12 simulated swaps

3 hours ago

Proposed

Allowed

Warnings

Blocked

DAO Operations Agent

DAO Agent

Active

Multi-sig coordination and governance execution

Connected Wallet0xd8dA...6045

ChainEthereum

Trust Score89/100

KINE Protection

Active

Last Action

Executed DAO proposal #47: 10 ETH grant distribution

5 days ago

Proposed

Allowed

Warnings

Blocked

Trust score / governance posture

How safely the agent is behaving

Agent Trust Scorecard

Comprehensive trust metrics for Clawdbot Treasury Agent

Overall Trust Score

Based on 6 key trust metrics

/100

High Trust - Agent is operating within safe parameters

Policy Compliance

Adherence to governance rules

Risk Tendency

Preference for low-risk actions

Approval Discipline

Requests within limits

Destination Hygiene

Uses verified addresses only

Protocol Reputation

Interacts with trusted protocols

Historical Blocks

Few blocked attempts

Trust Score Insights

Agent has maintained 94%+ policy compliance over the last 127 actions

Only 3 blocked actions out of 127 total proposals (2.4% block rate)

100% of proposed destinations are verified and allowlisted

5 warnings issued for high-value transactions requiring approval

Decision, token, and history summary

Governed outcome, economy, and evidence

Pending

Decision summary

Run governed task

Run Governed Task to generate a KINE decision, structured record, and browser-local history entry.

Token economy summary

Estimated avoided

Pending

Measured provider tokens

Not available yet

Reduction percentage

Pending

Cost avoided estimate

Pending

Structured record summary

Record ID

Run governed task to generate one

Decision ID

Pending

History entry

Pending

History size

0 records

Continue to Token Economy Sessions Supporting Evidence Feedback

Truth boundary

What this beta does not claim

Current beta does not store API keys.
Current beta does not give agents autonomous fund control.
Current beta does not broadcast transactions.
Current beta does not prove production agent safety.
BYOK remains a controlled beta path and does not grant autonomous wallet execution.

Advanced runtime diagnostics