I build verifiable execution systems for autonomous agents, programmable markets, and onchain payments.
An agent may propose an action, but it should not be able to authorize its own spending, exceed policy, grade its own result, or hide the evidence.
The missing layer in agent infrastructure is not another model or dashboard. It is an execution boundary that can answer four questions every time software tries to spend, trade, settle, or pay:
Is it allowed? → Is it still safe? → Did it work? → Can anyone verify that?
My projects turn that boundary into signed policies, independent verification, fail-closed state transitions, recovery paths, and portable evidence receipts.
Intent → Policy → Observe → Execute → Verify → Settle / Freeze → Receipt
Six focused products shipped between September 2 and September 14, 2026. They share a verification thesis, but each tests it against a different real-world execution surface.
| Product | The problem it solves | Public proof |
|---|---|---|
| OpenBell | Fair-execution control plane for 24/7 tokenized-stock markets on Solana; checks routes, liquidity, off-hours risk, corporate actions, and Token-2022 amounts before an agent trades. | Live Judge Demo · Mainnet AAPLx/Jupiter observation · signed policies · persistent task API · independently verifiable receipts |
| Allowance OS | Solana Mobile control center for subscriptions and autonomous-agent spending: approve once, enforce every charge, revoke anytime. | Public Demo · Android v0.18.0 · Devnet Delegated Settlement v2 · wallet-approved AlphaBrief settlement |
| ETHOnline AgentGuard | Policy-gated Agent-to-Agent commerce in which the buyer cannot authorize itself and the seller cannot grade its own output. | Public Demo · The Graph data · Privy authorization · Arc Testnet USDC · live VERIFIED / BLOCKED / FROZEN receipts |
| Proof-of-Calibration | Capital access for trading agents based on incremental value versus no-trade and benchmark counterfactuals—not raw profit. | Public Demo · proof-carrying capital permits · calibration passport · live read-only Bitget evidence |
| Monad AgentGuard | Bounded Agent-to-Agent execution and escrow with an independent verifier on Monad. | Public Demo · verified contracts · MetaMask Agent Wallet task · live Testnet settlement · parallel benchmarks |
| Binance AgentGuard | Accountable Binance Agentic execution without giving an AI agent unchecked exchange authority. | Judge Console · real bounded Spot execution evidence · independent order verification · tamper-evident receipts |
Agent Control Plane is the reusable TypeScript foundation underneath the venue-specific builds. It provides policy, risk, execution, verification, attestation, recovery, persistence, observability, and receipt primitives for autonomous economic activity.
| Layer | What it controls |
|---|---|
| Policy | permissions, budgets, targets, approvals, nonces, and time windows |
| Risk | exposure, duplication, price drift, slippage, liquidity, and circuit breakers |
| Execution | exchange, EVM, Solana, escrow, MCP, x402, and workflow adapters |
| Verification | independent outcome checks before release or settlement |
| Recovery | block, cancel, refund, freeze, retry, revoke, and escalation paths |
| Evidence | signed attestations, deterministic hashes, receipts, and audit history |
- Judge-operable: a reviewer can reach the core proof quickly without private credentials.
- Evidence-first: live data, onchain transactions, deterministic fixtures, and design-only boundaries are labeled separately.
- Fail-closed: missing routes, stale observations, invalid signatures, and unverifiable outcomes do not silently pass.
- Integration-minded: SDKs, APIs, schemas, examples, and reproducible checks are treated as product surfaces, not post-hackathon cleanup.
- Honest about maturity: Testnet, Devnet, unaudited, simulated, and production claims are never collapsed into the same label.
The current portfolio grew from earlier work in programmable wallets, multi-agent treasuries, dispute resolution, agent labor markets, privacy, and micropayments:
- AgentPay — programmable wallets and payment rails for AI agents.
- AgentBank — multi-agent autonomous treasury infrastructure.
- AgentCourt — verifiable execution and decentralized dispute handling.
- KiteHive — onchain Agent-to-Agent labor-market experiment.
- Cadence — autonomous royalty settlement with USDC nanopayments.
- Tokenized stocks: OpenBell repository · live Demo
- Solana Mobile payments: Allowance OS repository · Android releases
- Agent commerce: ETHOnline AgentGuard · public Demo
- Reusable infrastructure: Agent Control Plane
- Build notes and launches: X / @0xCaptain888