TrustProof-Protocol

LangChain Integration

What It Does

wrapToolWithTrustProof(...) wraps a LangChain tool call and emits a TrustProof JWT per invocation.

For each call, it:

• executes the original tool
• hashes canonicalized input and output
• builds claims envelope fields
• signs with generate(...)
• optionally chains with prior proof in memory

Implementation:

• ../../packages/js/src/integrations/langchain.ts

Run the Example

pnpm --filter @trustproof/sdk example:langchain

Example script:

• ../../examples/integrations/langchain/tool_call_proof.mjs

Minimal Usage

import { wrapToolWithTrustProof } from "@trustproof/sdk";

const wrapped = wrapToolWithTrustProof(tool, {
  privateKeyPem,
  subject: { type: "agent", id: "agent_demo_1" },
  action: "agent.tool.payout.quote",
  resource: { type: "tool", id: "payout.quote" },
  policy: { policy_v: "v0", scopes: ["tools:invoke"], constraints: {} }
});

const { output, proof_jwt } = await wrapped.invoke(input);

Naming Convention

Use stable, parseable action names:

• agent.tool.<domain>.<operation>
• examples: agent.tool.payout.quote, agent.tool.payout.initiate

This site is open source. Improve this page.