AI orchestration + project scaffolding for Stellar. Six agents write, verify, and deploy contracts, frontends, and payment APIs — no context-switching. The CLI bootstraps the project. The Skill builds it. Use either, or both.
Copilot and ChatGPT don't know contracts must be #![no_std], or that x402 needs OZ Channels and CAIP-2 network IDs. This framework embeds that domain knowledge into 10 skills, 6 agents, and 5 eval files.
10 bundled knowledge skills (soroban-sdk, Wallets Kit, x402, Groth16, SEPs, MCP, etc.) — loaded on demand, not dumped into context.
Not a chat. The kernel builds a work graph per task — which agents, in what order, sharing what state — and verifies each node before proceeding.
Every output checked against structured pass/fail evals. Fail → retry with corrective context (max 3). Pass → hand off to next node. No blind trust.
The CLI scaffolds a production-ready monorepo (contracts + frontend + backend + CI/CD). The Skill activates AI agents inside it. Together they're a complete workflow.
Six agent nodes, each owning a zone with persistent context and defined edges. Stable structure that persists across sessions.
The kernel is not a runtime — it is a structured prompt that tells the AI how to organize its own work. It generates a work graph for every task.
6 agent nodes, each with zone ownership, persistent context, and edge definitions. Stable across sessions.
Per-task dynamic wiring: sequential, parallel, conditional, fan-out, fan-in. Determined by data dependencies, not hardcoded order.
Each node gets intent + context + tools. Returns output + state delta + verifier result. No node writes code outside its zone.
Graph rewrites itself: spawn nodes mid-task, reroute on failure, collapse on early convergence, reorder on priority shift.
After each node, run its verifier. Pass → proceed. Fail → retry with corrective context (max 3) → reroute to fallback → escalate.
Shared state (contract IDs, deploy records, .env) travels along edges. Nodes never rediscover what sibling nodes already computed.
Two entry points, one framework. Here's how to choose.
Adds 6 AI agents to your sessions. The graph engine activates automatically — describe what you want in natural language.
Generates a production-ready monorepo: contracts, frontend, backend, CI/CD, and agent files. The CLI auto-installs the Skill too.
Six specialist agents, 10 domain skills, 5 eval files. From contract to deployment in minutes.
Scaffold Rust/Soroban contracts with build, test, and deploy workflows — testnet gated.
Generate Next.js apps pre-integrated with Stellar Wallets Kit, hooks, and tx flows.
Build API servers and indexers that query Stellar RPC and Horizon with x402 middleware.
Monetize APIs with HTTP 402 and Stellar USDC — zero-XLM clients via OZ Channels.
Integrate Groth16, Circom, and Noir verifiers into Stellar contracts via BLS12-381.
Every agent output verified against structured evals with max 3 retries. No blind trust.
Each agent is a structured prompt configuration (not a running process). They communicate through a defined graph — data flows along edges, not through the kernel.
From empty directory to deployed dApp — the graph engine handles routing, verification, and knowledge graphing.
Pick your entry point — Skill for AI orchestration, CLI for project scaffolding, or both.
Tell the graph engine what to build. It generates a work graph and routes to the right agents.
Agents write code, evals verify it, knowledge graph maps the project. Deploy with one command.
/deploy . testnet