AICorp
We fired the CFO. AI didn't want equity.
About
The problem it solves **Financial intelligence is slow, expensive, and locked behind human intermediaries.** A smart contract audit costs up to $200,000. A VC due diligence report takes two weeks. A treasury health check requires a full-time CFO. None of this makes sense when value moves in seconds and anyone can deploy a protocol overnight. **AICorp makes all of it instant, autonomous, and cheap.** --- ### ⚡ What people use it for | Task | Old Way | With AICorp | |--- |---|---| | 🏛️ Protocol treasury health check | Hire a CFO, wait days | 30 seconds, $0.25 | | 📈 Investment research brief | $300k/yr hedge fund analyst | Instant, $0.20 | | 🔍 VC-style due diligence | 2-week associate report | Instant, $0.30 | | 🛡️ Smart contract audit | $50k–$200k, months wait | Instant, $0.35 | No sign-ups. No invoices. No humans. Just submit a request, pay in USDC, get a structured report — with a blockchain receipt. Use of AI tools and agents AICorp is entirely built around autonomous AI agents — every piece of AI is purposeful, domain-specific, and economically accountable. --- ### 🏛️ Manager Agent — The CEO Handles all incoming requests, enforces **x402 payment collection**, verifies **treasury solvency** before every dispatch, routes to the right specialist, and logs every financial event on-chain. Holds an **ERC-8004 on-chain identity** on Kite Testnet. --- ### 🤖 Four Specialist Worker Agents Each agent follows the same autonomous loop: receives dispatch → collects **USDC payroll on Base Sepolia** → builds a domain-specific prompt → runs **inference on 0G Labs' decentralized compute** → returns a structured report → reports compute cost back as OPEX. | Agent | Domain | |---|---| | 📊 CFO Agent | Financial health, runway, risk classification | | 📈 Alpha Analyst | Investment thesis, signal extraction, verdict | | 🔍 DD Agent | Due diligence, red flag detection, competitive analysis | | 🛡️ Auditor Agent | Vulnerability identification, severity scoring | All agents use **qwen/qwen-2.5-7b-instruct via @0glabs/0g-serving-broker** for inference. Every agent's revenue, payroll, and compute cost is tracked individually — the system knows its own margin per agent, per request, in real time.
How it's made
Challenges we ran into ### 🔄 Nonce collisions on concurrent blockchain transactions The Manager Agent was sending treasury and payroll transactions in rapid succession from the same wallet, causing **replacement fee too low** errors. Fixed by centralizing the ethers.js provider and wrapping the signer with **NonceManager** to serialize nonce assignment. --- ### 🧩 ESM / CommonJS crash with the 0G SDK The 0G serving broker package crashed on Railway's default Node.js 18 with a **named export not found** syntax error. Fixed by pinning **Node.js 22** in nixpacks.toml to match the local environment where it worked fine. --- ### 📦 pnpm lockfile failing on Railway Railway expected a **pnpm-lock.yaml** but it was gitignored, breaking every build. Corepack also failed due to a key signature error in the Nix environment. Fixed by generating the lockfile locally, committing it, and using **npx pnpm@10** in the install phase. --- ### 🎨 Animated layers blocking interactive UI elements After adding glow effects and animated backgrounds, buttons and inputs stopped responding to clicks — decorative layers were sitting on top of them in the z-index stack. Fixed with **pointer-events: none** on all decorative elements and **isolation: isolate** on card components.