AI Agent Layer
AsiliChain deploys on-chain AI agents as a native layer of the protocol. These agents are not optional tools or background scripts. They are registered economic participants on Mantle Network, operating within the ERC-8004 standard, with verifiable on-chain identities, portable reputations, and cryptographically auditable decision records.
Every agent action is recorded permanently on Mantle. Every decision is auditable by regulators, MFIs, and cooperative managers without requiring trust in AsiliChain.
Why an Agent Layer
Section titled “Why an Agent Layer”The AsiliChain smart contracts handle economic logic: minting, lending, repayment, and fee collection. They do this correctly but without context. They cannot detect that a moisture reading is three standard deviations outside normal for that cooperative’s history. They cannot notice that a farm polygon has shifted between registrations. They cannot recommend forbearance before a drought causes defaults.
The AI agent layer adds contextual intelligence to the protocol without touching the on-chain security model. Agents read contract state, external data feeds, and historical patterns. They recommend or execute actions. Every recommendation is recorded on-chain before execution. No agent can modify a contract directly without going through the same access control roles that humans use.
The On-Chain Standard: ERC-8004
Section titled “The On-Chain Standard: ERC-8004”AsiliChain agents are registered under ERC-8004, the Ethereum standard for trustless agents deployed on Mantle mainnet in February 2026. ERC-8004 provides three on-chain registries that every AsiliChain agent uses:
Identity Registry Every agent receives a unique NFT-based on-chain identity. The AsiliChain LendingVault Risk Monitor, the USSD Session Agent, and the DDS Anomaly Detector are each separately identifiable, discoverable, and auditable on Mantle.
Reputation Registry Every agent builds a portable track record. Correct risk calls, accurate anomaly detections, and successful session completions are recorded against each agent’s identity. MFIs evaluating the protocol can query an agent’s performance history before trusting its recommendations.
Validation Registry Cryptographic proof of work completed. When the Risk Monitor recommends forbearance, the recommendation is logged with a proof that the agent evaluated the correct inputs. The recommendation is not erasable and cannot be backdated.
ERC-8004 is backwards-compatible with MCP (Model Context Protocol), A2A (Agent-to-Agent) communication, and the x402 payment standard. AsiliChain agents expose MCP endpoints for external agent integration.
Four Deployed Agents
Section titled “Four Deployed Agents”AsiliChain deploys four agents at specific points in the protocol where contextual intelligence creates measurable value. Each one is documented in its own page.
| Agent | Deployment point | ERC-8004 category |
|---|---|---|
| LendingVault Risk Monitor | Between Chainlink feed and LendingVault | RWA Coordination Agent |
| USSD Session Agent | Between Africa’s Talking and AsiliChain API | RWA Coordination Agent |
| DDS Anomaly Detector | Between TraceLog GRADED event and DDS pipeline | RWA Coordination Agent |
| Cooperative Intelligence Agent | Cooperative dashboard data layer | Financial Strategy Agent |
Phase 1 Implementation (API Layer)
Section titled “Phase 1 Implementation (API Layer)”Two agents are currently implemented in the API layer under lib/agents/:
LendingVault Risk Monitor
Section titled “LendingVault Risk Monitor”| Field | Value |
|---|---|
| Registry key | risk-monitor |
| Agent ID | asilichain-risk-monitor |
| Capabilities | LTV calculation, portfolio risk aggregation, price feed staleness detection, Hedera HCS audit trail |
| Trigger | scheduled — runs every 15 minutes |
| HCS event type | RISK_MONITOR_REPORT |
| Source | lib/agents/risk-monitor.ts, app/api/agents/risk-monitor/route.ts |
Reads loan positions from LendingVault, computes current LTV using pricePerKgBase, batch weight, and grade multiplier. Each active loan is classified as healthy (<80% maxLTV), warning (80–100%), or critical (≥100%). Portfolio-level statistics are aggregated and written to Hedera HCS.
DDS Anomaly Detector
Section titled “DDS Anomaly Detector”| Field | Value |
|---|---|
| Registry key | anomaly-detector |
| Agent ID | asilichain-anomaly-detector |
| Capabilities | Missing purchase order detection, stale export detection, weight outlier detection, stage distribution analysis, Hedera HCS audit trail |
| Trigger | event — fires on BATCH_SUBMITTED |
| HCS event type | ANOMALY_SCAN_COMPLETE |
| Source | lib/agents/anomaly-detector.ts, app/api/agents/anomaly-detector/route.ts |
Scans all batch tokens via multicall, reads stages from TraceLog and batchToActiveOrder from PurchaseOrder. Detects stalled exports (72h threshold), missing buyer commitments on WAREHOUSED/COMMITTED batches, missing grades, weight outliers, and stage distribution imbalances.
ERC-8004 on-chain registration completed on Mantle Sepolia (Phase 1). Both risk-monitor (agentId=0) and anomaly-detector (agentId=1) are registered on the IdentityRegistry at 0x62a6b58f8c3625F0c5f46D6C86A65595AA769C89. See the Architecture page for full deployment details.