Sivan AI

Inspiration

AI agents inside web browsers (ChatGPT, Google Gemini in Chrome, Claude) are rapidly advancing from passive text generators to active digital workers capable of negotiating deals, booking services, and coordinating freelance work. However, when browser agents attempt commercial transactions, they hit a major bottleneck:

1. UI Fragility: Scraping DOM elements or simulating blind clicks often fails when layouts shift.
2. Direct Transfer Risk: Upfront payments carry severe counterparty risk with zero delivery guarantees.
3. Disconnected Financial Rails: Agents lack unified multi-chain settlement capabilities and cannot bridge crypto assets into local fiat banking.

Sivan AI was built to solve this by providing the first autonomous, multi-chain settlement and Service Agreement layer for browser AI agents using the open W3C WebMCP standard.

What it does

Sivan AI enables AI agents in browsers to negotiate, draft, fund, and complete commercial Service Agreements with milestone protection.

By registering typed financial tools on document.modelContext, web pages allow browser agents to interact with Sivan AI natively:

• Create Milestone Agreements: The agent drafts clear deliverables and milestone terms.
• Query Spendable Balances: The agent checks available USDC balances across Solana, Base, Stellar, Celo, and BSC.
• Human-in-the-Loop Safety: Every funding action and milestone payout requires on-screen human approval via interactive review cards before any blockchain transaction moves.
• Instant Local Fiat Settlement: Settled USDC can be off-ramped directly into local bank accounts (e.g. Nigerian Naira via instant NGN bank rails).

How we built it

• WebMCP Client Layer: Implemented in TypeScript using official webmcp-types declarations, registering document.modelContext tools on https://staging.sivantech.online.
• Multi-Chain Core: Node.js and Fastify microservices orchestrating Privy embedded wallets, Stellar Horizon RPC, Solana web3.js, and EVM adapters.
• Edge Gateway: Cloudflare Workers (sivan-gateway) managing edge routing, CORS, and Model Context Protocol streaming.
• Cloud Backend: High-availability microservices deployed on Render with PostgreSQL persistence on Neon.

Challenges we overcame

1. Multi-Chain Fan-Out Latency: Grouped multi-chain RPC reads into deduplicated queries, preventing public RPC rate limits and timeouts from blocking balance rendering.
2. Agent Security Boundaries: Designed a strict human-in-the-loop authorization modal so autonomous agents can never drain or move funds without explicit user confirmation.
3. Cross-Chain Key Unification: Unified distinct cryptographic wallet families (ed25519 Solana, secp256k1 EVM, Stellar StrKey) under a single cohesive WebMCP tool schema.

Accomplishments that we are proud of

• Delivering a fully functional, live multi-chain settlement platform on staging (https://staging.sivantech.online) and Telegram AI (@SivanStaging_Bot).
• Achieving sub-second Solana and Base transaction broadcasts with verifiable block explorer receipts.
• Creating an open-source, standards-compliant WebMCP tool provider available on GitHub (https://github.com/Sivan-Technologies/Sivan).

What we learned

WebMCP fundamentally transforms web applications from human-only interfaces into dual-purpose environments where humans and AI agents collaborate seamlessly without breaking authentication or state boundaries.

What is next for Sivan AI

• Extending WebMCP tools to background service workers for asynchronous milestone monitoring.
• Enabling automated recurring service agreements and escrow-free micro-payouts.
• Onboarding global freelance marketplaces to Sivan AI agentic settlement rails.

Built With

• ai-agents
• base
• bsc
• celo
• chatgpt
• cloudflare-workers
• fastify
• node.js
• postgresql
• privy
• react
• render
• solana
• stellar
• typescript
• usdc
• w3c
• webmcp