Documentation
ProofLedger is model-independent trust infrastructure for AI agents: verifiable identities, model provenance, a tamper-evident audit trail, live trust scores, runtime policy evaluation, verified outcomes, and replay — all of which survive model changes. Integrate with the TypeScript or Python SDK, the REST API, or an OpenTelemetry exporter.
Install
# TypeScript / JavaScript
npm install @proofledger/sdk
# Python (add [signing] for Ed25519 signed events)
pip install 'proofledger[signing]'Then create an API key in the dashboard under API Keys and follow the quickstart.
Core concepts
Agent identity
Ed25519 keypairs per agent. Events are signed at the edge and verified server-side.
Audit events
Append-only, per-agent SHA-256 hash chains. Tampering is detectable, always.
Model provenance
Which model executed, where, at what cost — with fallbacks in the verified timeline. Identity is permanent; models are replaceable.
Outcome verification
Did it work, how do we know, and was it worth it? Outcomes are never self-declared.
Trust score
Every agent starts at 100. Violations subtract, healthy behavior adds — with reasons.
Policy engine
allow · warn · block · require_approval — evaluated before the action is approved.
Runtime timeline
Every decision, tool call, and policy result in chronological order, per agent.
Replay
Reconstruct any workflow step by step from the audit log. Nothing re-executes.
MCP & tool verification
A verified registry for tools and MCP servers. Unknown tools are blocked by default.
Alerts & webhooks
High-severity security events email your team automatically.
OpenTelemetry
Point an OTLP exporter at ProofLedger — existing traces become verifiable trust data.