VaultGraph is a trust and verification platform for AI agents. Your backend hashes sensitive context locally, signs a JobReceipt, and submits it to VaultGraph Platform. VaultGraph verifies the signature, stores the receipt, and surfaces audit history and trust scores for the agent deployment.
This repo contains example implementations of the VaultGraph SDK in various contexts, from a minimal direct integration to more complex LangChain and Vercel AI SDK agents.
- vendor-app: Minimal direct SDK flow that signs, verifies, and submits a receipt.
- ai-sdk-agent: Vercel AI SDK integration using
@vaultgraph/sdk/ai. - langchain-agent: LangChain.js integration using
@vaultgraph/sdk/langchain. - cookbook/github-triage: LangChain + LangGraph issue triage agent that submits one receipt per GitHub issue.
- mcp-server: MCP server minimal test and Claude Desktop config using
@vaultgraph/mcp-server.
This workspace includes a VS Code MCP configuration in .vscode/mcp.json. If your editor supports MCP tools, the following servers are available:
VaultGraph Docs: Connects tohttps://vaultgraph.com/docs/mcpso your agent can search and read the VaultGraph documentation directly from the official docs source.VaultGraph Receipts: Starts@vaultgraph/mcp-serverwithnpxand loads credentials from the root.envfile. This gives agents a local MCP server for submitting VaultGraph JobReceipts from the examples in this repository.
- Create an account in the VaultGraph Platform.
- Create a vendor API key, an agent, and a deployment. The full flow is in the setup guide.
- Register the matching Ed25519 public key as an active signing key on that deployment.
- Copy
.env.exampleto.envand fill in real values.
cp .env.example .env
cd vendor-app
npm install
npm run startEach example loads ../.env, so one root environment file is enough for the whole repo.