> For the complete documentation index, see [llms.txt](https://parad0xlabs.gitbook.io/parad0xlabs-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://parad0xlabs.gitbook.io/parad0xlabs-docs/openclaw-skills-+-agent-flight-recorder/openclaw.md).
# What It Is
### The flight recorder for AI agents. Prove what your agent did — and what it didn't.
***
## The Problem
You're starting to let AI agents do real things — spend money, call APIs, move data, act on your behalf. So here's the uncomfortable question: **when something goes wrong, can you actually prove what the agent did?**
Today the answer is "trust the logs." But a log is just text in a database. It can be edited, deleted, or quietly trimmed — by an attacker, a careless employee, or the platform itself. If your agent overspends, leaks something, or touches what it shouldn't, you have no independent proof of the order of events. And you certainly can't prove a *negative* — that your agent never paid a blocked address, or that the record you're looking at wasn't quietly censored down to hide the bad part.
For one personal assistant, fine. For a business running hundreds of agents that move real money, that's a lawsuit waiting to happen.
***
## What OpenClaw Vault Is
**In plain words:** OpenClaw Vault is a flight recorder for AI agents. Like the black box on an aircraft, it records what your agent does into a sealed, tamper-evident trail — and writes a fingerprint of that trail to the Solana blockchain (a public ledger no single company controls), so nobody, not even you, can secretly rewrite history afterward.
Every action — every API call, every payment, every decision — is cryptographically chained to the one before it (each entry carries a fingerprint of the previous one). Change any single entry and the chain visibly breaks. The result is an evidence trail you can hand to an auditor, an insurer, a customer, or a court, and they can check it themselves — without trusting you, and without you having to reveal everything in it.
What makes it different from every other "AI audit log" is that it ties together three things nobody else combines:
* **Identity** — *which* agent did it (its Agent Passport).
* **Money** — *what it paid*, proven by the on-chain x402 receipt.
* **Action** — *what it actually did*, sealed in order.
> *"Every AI black box proves what the agent did. Ours also proves what it did **not** do — and that you weren't handed a censored version."*
> *"Steal the agent's signing key today, and you still can't forge, reorder, or delete a single payment it recorded yesterday."*
***
## Who Needs This
* **Businesses deploying agents that spend money or touch sensitive data** — you need a record that holds up when a customer, a regulator, or a lawyer asks "what happened?"
* **Compliance and dispute resolution** — settle "the agent did X" arguments with proof, not screenshots.
* **Insurers and anyone underwriting agent behavior** — price the risk against an evidence trail that can't be quietly edited.
* **Builders who want their agent's track record to be provable** — a reputation you can demonstrate, not just claim.
***
## Why This Is The First Product People Will Actually Use
The rest of Web0 is the protocol. **OpenClaw is the first product built on all of it at once** — it uses the payment rail (x402 receipts), the identity layer (Agent Passport), the proof layer (Solana anchoring), and our compression engine (Liquefy) to do one job a business will pay for on day one: *make an agent's behavior provable.*
It's already real: the evidence engine is shipped and open, it installs with one command, and the OpenClaw agent-memory skill alone has 250+ installs.
**Status:** Public Beta. The on-chain anchor is confirmation-level (not yet finality-grade), and we don't say "audited" — the proof is the chain and the on-chain fingerprint, not a label.
***
## What ships alongside it — the `openclaw-skills` monorepo
The Vault is the flagship, but it now lives in **`openclaw-skills`** ([repo](https://github.com/Parad0x-Labs/openclaw-skills), MIT) — the claw-family **skills monorepo**. These are drop-in modules for agents and agent runtimes, each usable on its own:
| Skill | What it does |
| -------------------- | --------------------------------------------------------------------------------- |
| **Vault** (evidence) | the flight recorder above — tamper-evident trail + Solana anchor |
| **x402-pay** | client side of the x402 rail — quote, pay, attach the receipt |
| **x402-gate** | server side — return HTTP 402, verify the receipt, unlock the resource |
| **context-capsule** | portable, compressed agent context/memory (the agent-memory skill, 250+ installs) |
| **MCP server** | exposes the skills above to any MCP-speaking agent or runtime |
They're kept as independent, no-cross-import modules, so you can take one without dragging in the rest. Release state varies per skill — some are on npm/PyPI, some are install-from-source — so each skill's own README in the repo is the source of truth (we don't call a package published until it is).
***
**Next:** [**How It Works**](/parad0xlabs-docs/openclaw-skills-+-agent-flight-recorder/how-it-works.md) **— the evidence spine, the hash chain, and how a vault gets anchored to Solana.**
---
# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.
## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.
Perform an HTTP GET request on the current page URL with the `ask` query parameter:
```
GET https://parad0xlabs.gitbook.io/parad0xlabs-docs/openclaw-skills-+-agent-flight-recorder/openclaw.md?ask=
```
The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.
Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.