The problem OpenBook solves

                                                                                                                                                                                     Bond issuance is a $140 trillion market that still runs on phone calls, spreadsheets, and blind trust; the lack of technology in our era of innovation makes no sense. 

When a corporation or government wants to raise money by issuing bonds, they hire a lead bank to run a "book-build," collecting bids from institutional investors and deciding who gets what. This process has three deep structural problems:

1. No privacy between investors

Every investor's bid, including their size, price, and implicitly, their desperation, is visible to the bank. The bank knows exactly who will pay more. This creates systematic information asymmetry, enables front-running, and has led to billions in regulatory fines for unfair allocation practices (spinning, laddering). Investors can't trust that each other's bids won't be used against them.

2. The allocation is a black box

Investors submit bids and receive allocations with zero ability to verify that the process was fair. Investors do not know if there was a central algorithm, if certain clients got special treatment, or if any unjust partitioning was done. By relying solely on the bank's word, the centralized allocation system is untrustworthy and dangerous.

3. Compliance is manual and fragile

Regulators receive after-the-fact reports that they have to trust. KYC (Know Your Customer) verification, a form of mandatory identity verification preventing fraud and terrorism, is repeated manually across deals. Audit trails live in spreadsheets that can be altered. Settlement failures are handled bilaterally with no on-chain record.

OpenBook replaces this with a cryptographically verifiable, privacy-preserving capital markets platform:

• Canton DAML enforces structural privacy. Each investor's bid is sealed and only visible to them and the lead bank. Other investors are cryptographically excluded, not just locked out by policy, ensuring that other parties do not gain asymmetric information.
• 0G Compute runs the allocation algorithm inside a Trusted Execution Environment and produces a verifiable proof. The regulator (and anyone) can confirm the algorithm ran correctly, tamper-free.
• Hedera HCS writes every event, including the bid submitted, book closed, allocated, and settled events, to an immutable timestamped audit log. Nothing can be backdated or altered, ensuring complete transparency and security.
• ADI Foundation attestations replace manual KYC with on-chain accreditation checks. An investor either has a valid attestation, or they don't, which is verified in seconds, rather than days.
• ERC-4337 smart accounts give institutional investors a programmable on-chain identity on ADI Network, enabling gasless, compliant participation without managing raw private keys.

The result is a future of bond deals that any regulator can audit, any investor can trust, and any bank can run, built with the rapidly innovating technology of blockchain, artificial intelligence, and security.

ADI Testnet Keys:

Smart Account Factory: 0xbdEd0D2bf404bdcBa897a74E6657f1f12e5C6fb6
Paymaster: 0x2910E325cf29dd912E3476B61ef12F49cb931096
Attestation Registry (KYC): 0xA7918D253764E42d60C3ce2010a34d5a1e7C1398
EntryPoint (ERC-4337): 0x0000000071727De22E5E9d8BAf0edAc6f37da032
Settlement Ledger: 0xadAD55f56C23cF8B1286A3419bFeed055F1aDcb0

Challenges we ran into

OpenBook touches ADI Foundation, Canton, 0G, Hedera, and QuickNode, each with their own SDKs, authentication models, testnets, and quirks. The hardest part wasn't any single integration; rather, it was keeping them all working simultaneously. Although in the end, we believe every chain and technology we used served a meaningful and necessary purpose, the struggles of integrating each one into a meaningful product was difficult.

Likewise, since standard investors would likely be unfamiliar with how blockchain and cryptocurrency wallets work, one of our goals was to enable judges and investors to log in and immediately participate: no faucets, no wallet setup, no friction. Achieving that on the ADI testnet with ERC-4337 smart accounts required rethinking the standard deployment flow. We moved smart account deployment server-side so that an operator wallet sponsors the gas, the account lands at the user's deterministic address, and the user's Privy wallet is immediately the owner, with zero prompts or prerequisites.

Use of AI tools and agents

We used two main AI agents in our project. The core AI agent in OpenBook is the bond allocation model running on 0G Compute, a decentralized compute network. When the Lead Bank closes the order book, the allocation job is submitted to 0G's inference network with the full set of investor bids, like quantity, limit price, and quality score. The model runs a multi-factor optimization that respects maximum investor concentration limits, prioritizes price discovery, and balances institutional demand.

What makes this meaningful beyond a standard LLM call is the cryptographic proof 0G returns alongside the result. It generates a receipt containing a model hash, input hash, and output hash, allowing anyone to verify that a specific, approved model processed the exact bids submitted and produced the allocation result shown.

Likewise, before any investor can submit a bid, they pass through an automated KYC compliance agent backed by ADI Foundation's on-chain attestation registry. The agent verifies that the investor's wallet holds a valid accredited investor attestation, checks it hasn't expired or been revoked, and gates bid submission accordingly. This replaces the traditional model where KYC is a manual, per-deal process handled by compliance teams over days, with a check that completes in seconds and produces a cryptographically verifiable result anyone on the network can inspect.

The two agents operate at different stages of the deal lifecycle. The KYC agent acts as the entry gate, while the 0G allocation agent acts as the decision engine. Once the book closes, it processes only verified bids and returns a tamper-proof allocation. Together, they enforce the two properties institutional capital markets require most: that participants are who they say they are, and that the process was followed exactly as specified.