Eight packages. One signed-inference stack.
Six MIT-licensed packages on npm, two on PyPI. Build agents that sign their own requests, pay per call in USDC on Solana or Base, verify any receipt end-to-end, and serve signed inference from any framework — Vercel AI SDK, Mastra, LangChain, MCP, Express, Hono, Next.js.
Pick the one that matches your role.
Use one, use all eight — they compose without forcing a stack on you. Each is independently published and versioned. Mix TypeScript and Python freely; they speak the same SIR v2 receipt format.
@vdm-nexus/sdk
Generate a fresh Ed25519 keypair (or load an existing one), sign requests, and call any signed-inference endpoint. The agent's public key IS their identity — no API keys, no rotating secrets.
Use it when Building an agent that calls Nexus or any SIR-compatible endpoint.
@vdm-nexus/x402
x402-native client for paying per inference call in USDC (Solana or Base) and verifying any returned receipt end-to-end. Five-check verifier: hashes, operator signature, on-chain settlement, payer match, and recipient anchor.
Use it when Building an agent that pays per call, OR verifying receipts you got from somewhere else.
@vdm-nexus/paywall
Drop-in middleware for Express, Hono, or Next.js. Wrap any handler; every paid call returns a signed Ed25519 receipt of what your handler produced. You become an SIR endpoint without writing the protocol layer.
Use it when Monetizing your own AI API with per-call USDC + cryptographic receipts.
@vdm-nexus/mcp
Model Context Protocol server that exposes signed inference as a tool. Drop it into your Claude Desktop, Cursor, or any MCP-aware client config — your agent immediately gets paid calls with verifiable receipts. Zero code change.
Use it when Plugging Nexus into a Claude Desktop / Cursor workflow without writing a custom agent.
@vdm-nexus/ai-sdk-provider
Drop-in provider for the Vercel AI SDK. Use Nexus inference inside generateText, streamText, and tool calls without changing your codebase. Every response carries a signed receipt accessible via the provider's metadata.
Use it when Already on the Vercel AI SDK and want signed-inference receipts under the hood.
@vdm-nexus/mastra-provider
Drop-in provider for Mastra. Wire signed inference into Mastra workflows and agents with a single import; receipts thread through each response for downstream verification.
Use it when Running Mastra and want signed-inference receipts on every model call.
vdm-nexus
Python equivalent of @vdm-nexus/sdk + @vdm-nexus/x402. Generate Ed25519 agents, sign requests, pay per call in USDC on Solana or Base, verify any receipt end-to-end. Same SIR v2 format as the TypeScript stack — mix languages freely.
Use it when Building Python agents that pay per call and verify receipts.
langchain-vdm-nexus
LangChain adapter for Python. Use Nexus signed inference inside LangChain chains, agents, and LCEL pipelines. Receipt metadata is preserved through the LangChain runnable interface so downstream tools can verify what each model call returned.
Use it when Building LangChain agents in Python and want signed-inference receipts on every model call.
One protocol, many entry points.
Same SIR v2 receipt format across all eight packages. TypeScript or Python, OpenAI-compatible endpoint or paywall middleware, Vercel AI SDK or Mastra or LangChain or MCP — pick the surface that fits your stack. Receipts are portable across all of them.
Agent (you) Nexus / any SIR endpoint
─────────── ───────────────────────────
@vdm-nexus/sdk @vdm-nexus/paywall
│ │
│ signed request │ signed receipt
└──────────────►────────┘
│
▼
@vdm-nexus/x402.verifyReceipt
│
(verify 5 checks: hashes,
signature, on-chain tx,
payer match, recipient)
Or skip the custom agent and use MCP:
Claude Desktop / Cursor
│
│ config: @vdm-nexus/mcp
└──────────────► NexusThree things we will not change.
MIT, open source
Every package is MIT-licensed and developed in the open at github.com/vdmnexus/vdmnexus. Fork it, embed it, ship it.
Tiny dependency surface
@vdm-nexus/sdk has two runtime deps. The whole chain is auditable in an afternoon. No mystery transitive load.
Spec is public
Receipts follow the SIR v2 spec. Any package — ours or someone else's — can produce or verify them. The protocol matters, not the library.
Every package above runs a real call in the playground.