Make a wallet
legible.
Agent Passport evaluates observable Algorand activity and relationships, then returns deterministic application-facing signals. These docs explain the implementation and its boundaries.
Start here
Decision model
- Trust scoring→
Five weighted sub-scores, inactivity adjustment, risk bands, and fresh-wallet behavior.
- Delegation→
Depth, sponsor quality, amount scoring, bounded traversal, and registry boxes.
- Sybil indicators→
Eleven weighted wallet-history and graph inputs with heuristic limits.
- Credit & underwriting→
Capacity estimation, four underwriting factors, approval gates, and exposure safeguards.
- Passport document→
Schema fields, data sources, and what the SHA-256 checksum covers.
API and SDKs
- Endpoint reference→
Methods, paths, inputs, response semantics, and operational side effects.
- Authentication→
Conditional HMAC signing, timestamp window, and canonical request fields.
- Idempotency→
Required mutating-request header and process-local 24-hour records.
- x402→
Optional facilitator-backed payment requirements and verification flow.
- TypeScript and Python→
Typed clients, retries, errors, and callback surfaces.
Operate it
- Architecture→
Read lifecycle, mutation lifecycle, Algorand boundaries, and caches.
- Deployment→
Node, Docker, testnet/mainnet configuration, and readiness behavior.
- Scaling→
Which stores are local and what must move to shared storage for replicas.
- Observability→
Prometheus metrics, active alert rules, Grafana panels, and SLO profiles.
- Security limits→
What observable wallet signals can and cannot establish.
Signals are evidence, not identity.
A trust score does not prove a human identity, replace KYC/AML obligations, guarantee repayment, or make a counterparty safe. Sybil analysis is heuristic. New wallets, stale activity, Algorand availability, provider configuration, and local operational state all affect the result.