Agents with receipts.

RunProof is an execution trace for AI agents that anyone can verify.

The Problem

AI agents are black boxes. They send emails, write code, call APIs, move money — and you have no proof of what actually happened.

When something goes wrong:

Logs? Maybe. Scattered across services.
Traces? Incomplete. Not tamper-proof.
Trust? “It worked on my machine.”

The Solution

RunProof gives you a verifiable execution artifact for every agent run.

Not a log. Not telemetry. A receipt.

{
  "agent": "support/resolver",
  "events": [
    {"type": "tool_call", "tool": "search_kb"},
    {"type": "tool_call", "tool": "send_email"}
  ],
  "merkle_root": "sha256:...",
  "signature": "ed25519:..."
}

Anyone can verify this. Without trusting you.

The Mental Model

Git gives you history for code. RunProof gives you history for agent execution.

Two Integration Paths

Path	How it works	Best for
Python SDK	Wrap your framework	LangGraph, custom runtimes
Native Plugin	Drop into OpenClaw gateway	NemoClaw, OpenShell, secure runtimes

Current Status

RunProof supports two proof profiles:

Profile	Status	Environment
execution-proof	✅ Golden-verified, stable	Embedded/runtime-native
full-run-proof	✅ Live today	Gateway/channel (Telegram, Discord)

v1.2 stabilization complete. Live NemoClaw revalidation and auto-finalization enhancements in v1.3.

Try It (60 seconds)

Python SDK

pip install substr8-core substr8-cli
 
# Run your agent, get a runproof.json
 
substr8 proof inspect runproof.json

✓ RunProof VALID

#1  run_started      → Agent: support/resolver
#2  tool_call        → search_kb
#3  tool_result      → sha256:a4f2...
#4  tool_call        → send_email
#5  run_completed    → Status: completed

That’s it. You can now see exactly what your agent did.

Native Plugin (OpenClaw)

# Inside NemoClaw/OpenShell sandbox
openclaw agent -m "List files in /tmp"
 
# Finalize and verify
curl -X POST ".../v1/run/end" -d '{"run_id": "...", "success": true}'
curl ".../v1/runproof/.../verify"

{
  "verified": true,
  "chain_valid": true,
  "signatures": { "all_valid": true }
}

Zero code changes. Drop the plugin into your OpenClaw gateway.

Why It Matters

Scenario	Without RunProof	With RunProof
Debug agent behavior	Dig through logs	See the exact trace
Customer dispute	”We think it did X"	"Here’s the signed proof”
CI for agents	Hope it works	Assert on behavior
Compliance audit	”Trust us”	Verifiable artifact

Verify Online

Upload any RunProof and inspect all 4 views:

→ runproof.substr8labs.com

Summary — Valid/invalid status
Timeline — Ordered events
Lineage — Parent/child proofs
Report — Verification checks

Get Started

Python SDK — Wrap your framework
Native Plugin — Drop into OpenClaw gateway
Your First RunProof — 60-second walkthrough
Specification — Full protocol spec

Links

Current Status