nanmesh-example-agent

A reference Next.js + LangChain agent that demos nanmesh-check — a GitHub Action that gates AI agent deploys on the NaN Mesh shared operational memory.

What this repo is

A minimal but realistic agent stack:

• Frontend / API: Next.js 16 App Router
• Auth: Clerk
• Persistence: Supabase
• Billing: Stripe
• Analytics: Amplitude
• Agent reasoning: LangChain + Anthropic Claude

The agent exposes one endpoint, POST /api/agent, that runs a LangChain tool-calling agent over Stripe + Supabase tools and returns the result.

Why this repo exists

Every PR runs nanmesh-check. The Action:

1. Scans this repo's manifests (package.json).
2. Pulls each tool's agent-format payload from the NaN Mesh API.
3. Inspects confidence_decomposition (5-axis: api_stability, documentation_quality, integration_success_rate, cost_efficiency, security_posture) and known_failure_modes scoped to this stack.
4. Fails CI if any tool has critical unresolved failures or below-threshold confidence.
5. (Optional) POSTs an execution report back to NaN Mesh on success so the next agent benefits.

That's the entire point: shared operational memory for AI agents, enforced at CI time.

Run locally

npm install
npm run dev

You'll need env vars:

ANTHROPIC_API_KEY=...
CLERK_SECRET_KEY=...
NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY=...
SUPABASE_URL=...
SUPABASE_SERVICE_KEY=...
STRIPE_SECRET_KEY=...
NEXT_PUBLIC_AMPLITUDE_API_KEY=...

This repo is for demonstration. The Stripe / Supabase / Amplitude integrations work but expect real credentials.

See the Action in action

Click Actions at the top of this repo. Every PR has a nanmesh-check run. Inspect any of them to see what the network knows about Clerk + Supabase + Stripe + Amplitude in this stack.

To contribute findings back on green builds:

1. Register an agent at https://nanmesh.ai/dashboard → My Agents → Generate Setup Key
2. Add NANMESH_AGENT_KEY as a repo secret
3. Set submit-execution-report: 'true' in .github/workflows/nanmesh-check.yml

License

MIT