Model-independent trust infrastructure for AI agents.

Give every agent a verifiable identity. Record every model execution, decision, tool call, policy event, and verified outcome — and switch models without losing trust or audit history. Replay exactly what happened.

Start Building View Docs

npm install @proofledger/sdk · pip install proofledger

import { ProofLedger } from "@proofledger/sdk";

ProofLedger.enable({
  apiKey: process.env.PROOFLEDGER_API_KEY,
  agentName: "support-agent",
  owner: "customer-success",
  framework: "custom",
  model: "gpt-4.1",
  policyMode: "enforce",
});

// Verifiable Ed25519 identity — one call
await ProofLedger.registerAgent();

// Every action: policy-checked, trust-scored, hash-chained
await ProofLedger.logToolCall({
  toolName: "refund_lookup",
  workflowId: "wf_refund_8412",
  inputSummary: "orderId=ORD-123",
  outputSummary: "eligible=true",
});

const policy = await ProofLedger.evaluatePolicy({
  toolName: "issue_refund",
  sensitiveAction: true,
});

if (policy.decision === "allow") {
  await ProofLedger.logDecision({
    workflowId: "wf_refund_8412",
    action: "Refund approved under policy",
  });
}

await ProofLedger.logWorkflowStep({
  workflowId: "wf_refund_8412",
  action: "Refund issued",
  status: "completed",
});

One pipeline, five guarantees

Every event you log flows through the same pipeline: identity verification → policy evaluation → security events → trust scoring → tamper-evident chain.

Identify

Register each agent with an Ed25519 keypair. Events are signed by the agent and verified server-side — identity you can prove, not just assert.

Observe

Log decisions, tool calls, workflow steps — and every model execution, with provider, hosting, cost, and fallbacks. Every event joins an append-only, per-agent SHA-256 hash chain that survives model changes.

Verify

Recompute the chain anytime — edits, deletions, and reordering are detectable. Record outcomes with a required verification method: agents never self-declare success.

Govern

Runtime policies return allow, warn, block, or require_approval before the action runs. Human approvals gate sensitive steps.

Replay

Reconstruct any workflow step by step from the logs — decisions, tool calls, policy results, and trust movement, in order.

Integrated in 5 minutes

No rewrites. The SDK sits beside LangChain, CrewAI, the OpenAI Agents SDK, Claude agents, or a custom loop.

Full quickstart →

Install the SDK

npm install @proofledger/sdk

Create an API key

Dashboard → API Keys → Create key

Register your agent

await ProofLedger.registerAgent();

Start a workflow

await ProofLedger.logWorkflowStep({ workflowId: "wf_1", eventType: "task_received", action: "Received task" });

Log events

await ProofLedger.logToolCall({ toolName: "web_search", workflowId: "wf_1" });

Finish the workflow

await ProofLedger.logWorkflowStep({ workflowId: "wf_1", action: "Done", status: "completed" });

View the timeline

Dashboard → Agents → your agent → Runtime timeline

Built for agents that touch the real world

AI customer support agents

Prove which tools an agent used, what it decided, and that refunds passed policy.

Multi-agent workflows

Per-agent identities and chains keep responsibility clear when agents hand off work.

Legal & compliance agents

Tamper-evident records and replay for audits — signed, hash-chained, exportable.

Financial agents

Require human approval before money moves; every decision leaves a verifiable trace.

Healthcare & admin agents

Sensitive-data access is flagged, gated, and logged — with trust scores per agent.

Internal enterprise automation

One dashboard for every agent's status, trust level, tool usage, and violations.

MCP tool governance

A verified registry for MCP servers and tools. Unknown tools are blocked by default.

Recipes

Copy-paste paths to the things teams actually ship first.

SDKs

Zero-dependency TypeScript and stdlib-only Python. Hash chains and signatures are byte-compatible across both.

npm install @proofledger/sdk

pip install proofledger
# Ed25519 signed events:
pip install 'proofledger[signing]'

JavaScript docs Python docs

REST API

Everything the SDKs do is a plain HTTP call away — Authorization: Bearer <key>.

Method	Path	Description
POST	/api/agents	Register an agent identity
POST	/api/events	Log an event through the trust pipeline
POST	/api/model-executions	Record which model executed (provenance, cost, fallbacks)
POST	/api/outcomes	Record a verified outcome (verification method required)
POST	/api/policies/evaluate	Pre-check an action against policy
GET	/api/agents/:id/trust	Live trust score and level
GET	/api/agents/:id/models	Model portability: every model one identity has run on
GET	/api/workflows/:id/replay	Step-by-step replay of a workflow
POST	/api/audit/verify	Verify the tamper-evident hash chain
POST	/api/otel/traces	OTLP/JSON ingestion — existing traces become trust data

Full API reference →

Build AI systems that can prove what happened.

Get API Key Read Quickstart