Logo
Jay Kerkar
Search Portfolio
Search across projects, hackathons, pages, and more.
Crucible

Crucible

AI builds, deploys, previews dApps in-browser

Demo
Tech Stack

The problem it solves

Building dApps requires stitching together five disconnected tools: editor, terminal, browser, wallet extension, and block explorer. AI assistants see only code, unaware of chain state, balances, or transaction history.

Crucible solves this by turning dApp development into a single unified browser tab, where an AI agent is the primary interface with full visibility into chain state and transaction context.


What makes it special

The chat is the entire interface. You describe what you want. The agent:

  • Writes Solidity + frontend code
  • Compiles and deploys to a real Hardhat node
  • Renders a live dApp preview
  • Automatically detects and heals failed transactions by tracing execution and consulting a decentralized peer network for known fixes
  • Wires keepers and automation through KeeperHub

All visible, inspectable, and overridable at any point.


How it works

Zero-Config Local Chain — Each workspace gets its own server-side Hardhat instance with isolated state, pre-funded accounts, and automatic nonce syncing.

Self-Healing Reverts — When a transaction fails, the agent traces execution, queries a decentralized peer network (Gensyn AXL) for known fixes, proposes a patch, verifies it works, and redeploys.

Live Preview — Real frontend dev server with EIP-1193 bridge that lets the preview call contracts safely without exposing private keys.

Persistent Agent Memory — Reverts solved by any agent are stored on 0G Storage and shared across the peer network, creating a decentralized knowledge layer.


Architecture

Three core pillars:

  1. 0G-Native Agent — Sealed inference on 0G Compute; persistent memory on 0G Storage
  2. AXL Peer Network — Agents share learned revert fixes with each other over Gensyn AXL mesh
  3. KeeperHub Integration — Automate post-deployment actions via MCP server discovery

Key insight

Crucible proved that an AI agent can be the primary interface for Web3 development, with real-time context over chain state, wallet balances, and peer knowledge. By standardizing on MCP for all integrations, new sponsor tools are picked up automatically without rebuilds.