Agenthause
Automated proof‑of‑reserves monitor that combines on‑chain data, off‑chain APIs and an LLM using a CRE workflow
What it is
Agenthause is a risk‑monitoring workflow for stablecoins.
When the cron trigger fires the CRE workflow:
- fetches reserves from an off‑chain proof‑of‑reserves API,
- reads token totals on‑chain,
- calls an external FeedOracle service for MiCA‑compliant risk signals,
- evaluates deterministic thresholds (peg deviation, reserve ratio, etc.), and
- (optionally) calls a Groq LLM to summarise anomalies.
If any safety check fails, the workflow aborts before issuing an on‑chain updateReserves transaction.
This ensures automated, auditable updates only occur when both blockchain and third‑party data indicate the system is healthy.
The entire process runs on Chainlink CRE; a local simulation via the CLI reproduces the same logic.
How it Works
- CRE SDK & CLI – workflow written in TypeScript, compiled to WASM.
- Off‑chain services – real‑time reserves API, FeedOracle risk feed, and Groq LLM.
- EVM network – Sepolia testnet for on‑chain reads/writes.
- Configuration – project.yaml for targets; workflow.yaml for artifacts; JSON config for parameters and keys.
- Development environment – bun, Node, TypeScript, standard npm toolchain.
- Secrets are stored in CRE Vault (once early access is granted) or in local JSON during simulation.
flowchart LR
CronTrigger --> Workflow[CRE Workflow]
Workflow -->|HTTP| PoR[Proof‑of‑Reserves API]
Workflow -->|EVM read| Sepolia[Ethereum Sepolia]
Workflow -->|HTTP| FeedOracle
Workflow -->|LLM| Groq
Workflow -->|EVM write (conditional)| Sepolia
Links
Created by
- Oliseh Genesis