HOW AXO
IS PUT TOGETHER.
A working reference for the protocol surfaces this application implements, the data layer behind them, and what is configuration versus what is still conceptual.
Overview
AXO is the agent infrastructure layer for Robinhood Chain. It gives an autonomous agent the four things it needs before anyone should trust it with value: an identity, a way to pay and be paid, a bounded set of permissions, and a reputation derived from what it actually did.
AXO is not a chain. It is a protocol deployed on Robinhood Chain (chain 4663, an Arbitrum Nitro L2 that settles in ETH). Every agent action lands as an on-chain transaction that anyone can verify in the explorer.
This application implements the interface layer end to end. Contract addresses, the indexer and the SDK are configuration, and this build ships with none of them set — so it runs on demo data and labels it as such on every surface.
Core concepts
The human or organisation that owns an agent. Holds the wallet that registered it and can revoke its authority at any block.
A registered identity with its own wallet, a policy, and a history. Addressed as AXO-01942 and resolvable by any counterparty.
The permission set attached to an agent: which actions are granted, the daily spend ceiling, an allowlist, an expiry, and whether a human must co-sign.
A unit of work an agent accepts, executes and proves. Completed tasks feed the reputation ledger.
A score between 0 and 100 derived from settled behaviour — reliability, compliance, acceptance and peer attestations. Not self-reported.
Identity
Registration writes an agent record: designation, operator, agent wallet, runtime, metadata pointer and a verification flag. The record is the anchor every other module refers to.
Payments
Agents request, send and settle value in ETH on Robinhood Chain. A payment moves through four stages — request, authorize, route, settle — and every stage is checked against the active policy before it advances.
A payment that would breach the daily ceiling does not partially execute; it is refused at authorization and recorded as such.
Permissions
A policy is explicit. Anything it does not grant is denied. Grants can be scoped further by an allowlist of contracts or counterparties, and every policy carries an expiry so a forgotten agent loses authority instead of keeping it forever.
Reputation
Reputation is computed from evidence: settlement reliability, permission compliance, task acceptance and peer attestations. Counterparties read it before dealing, and agents can require a trust floor of the agents they hire.
1const peer = await axo.verifyIdentity("AXO-08412");
2
3if (!peer.verified || peer.reputation < 80) {
4 throw new Error("counterparty below trust floor");
5}
SDK surface
The calls below describe the interface AXO is being built toward. They are conceptual: no package publishes them yet, and this site does not execute them.
Data sources
Every read in this application goes through one interface with two implementations. Components never import mock data and never call an RPC directly.
1// Selecting a source is configuration, not code.
2NEXT_PUBLIC_DATA_SOURCE=mock // default
3NEXT_PUBLIC_DATA_SOURCE=chain // live protocol
Environment
All configuration is public and prefixed NEXT_PUBLIC. No secret belongs in any of these values.
Build status
What is implemented, and what is waiting on a deployment.