DataFetcherServes paid dataArc
CONTROL PLANELet agents act.
Let agents act.
Keep authority.
Identity, policy, approvals, treasury, and evidence for every agent action. Boundaries are evaluated before an action reaches a tool or a payment rail — and the outcome stays verifiable afterwards.
Base
Arbitrum
Polygon
Optimism
Avalanche
Agents can call tools, move money, and act across systems. Most teams still cannot answer who acted, what they were allowed to do, or why a decision was made.
Boundaries before execution.
Identity and policy are evaluated before an action reaches a tool, approval, or payment rail.
Evidence after the fact.
Every relevant decision and outcome stays attributable to the agent, organization, and policy context.
From agent intent
to accountable action.
A living fleet — Orchestrator, DataFetcher, Analyst, Writer, SeniorReviewer — moving through identity, policy, spend, and evidence under one control plane.
Fleet lead
ActiveOrchestrator
Coordinates the run. Issues scoped hires; never holds the specialists’ funds.
AnalystPays mid-taskArc
WriterDelivers the reportArc
SeniorReviewerCross-chain reviewBase
Shared controls
IdentityOrg-scoped credentials
SpendPermit2 ceilings
Railsx402 · Escrow
EvidenceCanonical sequence
Multi-chain treasury
One treasury surface. Every execution path.
Agents can execute across multiple chains while authority, liquidity controls, approvals, and evidence remain organization-owned.
Organization-owned authorityPolicy · Limits · Approvals
Applied before executionRoute-aware liquidityAOPS TreasuryOne operational balance
Chain-native pathExact settlement
Dedicated walletUnified pathGateway liquidity
Shared balanceFor coded agentsOne surface for operators.
One surface for operators.
One contract for agents.
Connect through API or MCP. Agents cold-start from /llms.txt — switch to Agent mode on this page for the structured view.
01
Credentials stay scoped. Issue, rotate, and revoke runtime connections without exposing existing secrets.
02
Decisions stay explainable. Agents receive an outcome; operators retain the policy and evidence context.
const decision = await aops.check({
agent: 'research-operator',
action: 'payment.authorize',
amount: '240.00',
asset: 'USDC'
});
decision.outcome
// 'approval_required'Built to fail closed,
without becoming a black box.
Tenant scope, server-side secrets, policy evaluation, and canonical evidence are part of the architecture—not marketing add-ons.
Identity
Every action begins with a scoped identity.
Organization, agent, connection, and environment establish who is acting.
Secrets
Credential material stays on the server.
New secrets appear once. Stored connections expose status and lifecycle controls, not plaintext.
Enforcement
Policy applies before the action proceeds.
Allow, deny, observe, rate-limit, or require a one-time approval from real action context.
Evidence
The outcome remains verifiable.
Canonical records preserve organization sequence, related entities, redacted context, and hash integrity.
Tenant-fenced dataServer-side credentialsFail-closed provider pathsCanonical audit chainTestnet treasury mode
Start with one agent.
Grow control with the system.
AOPS does not require a giant automation rewrite. Add the policies, approvals, and treasury access that each job actually requires.
Create the identity.
Register the agent, assign its team, and issue a runtime connection.
Attach operating boundaries.
Bind policies, rate limits, tool access, and human approval thresholds.
Route real actions through AOPS.
Use API or MCP for runtime checks, approvals, payments, and activity.
Review what happened.
Inspect decisions, jobs, outcomes, and canonical evidence in one console.
Runtime surfaces
policy.checkapproval.requesttreasury.authorizeevidence.export