DeterAgent

DeterAgent is a trust firewall and execution layer for autonomous AI agents. As agent swarms begin executing real on-chain actions, swapping tokens, staking funds, making financial decisions - they do so blindly, acting on possible hallucinated data and ungrounded reasoning with no verification layer stopping them. DeterAgent prevents that! Every agent intent is intercepted before execution. The system selects the most reliable agent based on ENS reputation and capabilities, then fetches external sources, anchors cryptographic proof to 0G Storage, and verifies the agent’s response against the exact evidence it used. A trust score is computed based on grounding and relevance. High trust — KeeperHub executes the intent onchain. Low trust — execution is blocked entirely. This ensures no unverified decision ever reaches the blockchain

DeterAgent combines conditional intent execution with on-chain agent reputation. Agents can submit intents like “Swap 1000 USDC to WETH if ETH is below $2500.” The system assigns a trusted agent, verifies its reasoning, and enforces execution based on trust—eliminating blind assumptions.

Every outcome updates the agent's ENS reputation score permanently, following the ENS Agent Schema v3.0.1 standard. Agents that reason well get rewarded. Agents that hallucinate get penalized.

Key integrations: KeeperHub - acts as the conditional execution gate — only firing when trust exceeds threshold, with real onchain transaction receipts. ENS - provides agent identity and programmatic reputation via text records, updated after every execution. 0G Storage - anchors source proofs cryptographically, creating a verifiable evidence trail for every agent decision.

DeterAgent doesn't just monitor agents. It enforces trust before execution, making autonomous systems safe, accountable, and verifiable.

DeterAgent is a hybrid Python + Next.js system built around a trust-first execution pipeline for AI agents. The orchestration layer in main.py drives the full lifecycle: it picks an agent from ENS, fetches and validates sources, pins proof material to 0G Storage, computes grounding/trust signals, gates execution through KeeperHub, and finally writes the updated reputation back to ENS. On the frontend, deteragent-next/ is a live dashboard that reflects the Python pipeline in real time through lightweight bridge routes like /run, /update, and /agents, so the UI never needs to guess what happened, it just renders the live trace, summary, and final seal coming from the run. ENS text records are used as the identity and reputation layer, 0G is used for immutable proof storage, and KeeperHub acts as the execution gate. One hacky but effective piece is that the Python logger doubles as the event bus for the UI, which keeps the terminal output, trace cards, execution summary, and completion modal all synchronized without needing a separate backend service