Guide
How to prove your AI actually did the work
Short answer: To prove your AI actually did the work, have it run its steps through a tool layer that keeps a record of each call. Zambo does this: connect zambo.dev/mcp, and every tool call returns a verifiable receipt with a public link that anyone can open and check.
Why "Done!" isn't proof
AI assistants are good at sounding finished. A chat reply that says "I checked the price" or "I pulled the report" is only text. It doesn't show that a tool ran, what came back, or when it happened. If the AI skipped a step, guessed, or got an error it didn't mention, the chat looks the same.
What you need is a record that sits outside the chat: a note made when the tool actually ran, which you can look at later.
What a verifiable receipt is (and isn't)
When your AI calls a tool through Zambo, the call gets a verifiable receipt. It's saved at a public page (zambo.dev/run/<id>) with a verify link (/api/v2/receipt/<id>/verify). The receipt records which tool ran, what it returned, and when. It's written in the open AER-1 format.
Be clear about the limits. A receipt shows the saved output hasn't changed since the call. It doesn't show the output is correct. It proves the step happened and what came back. Whether the answer is any good is still your call.
Step by step: try it in about a minute
- Connect. Paste
https://zambo.dev/mcpinto your AI's connector or MCP settings. No key, no signup. - Make one real call. Ask: "Use zambo live_price for BTC." The AI calls
live_priceand replies with the price plus a receipt link. - Open the receipt. Click the
/run/<id>link. You'll see the tool, the input, the output and the time. - Verify it. Use the verify link on the page. It checks that the saved output still matches what was recorded.
- Do real work the same way. Ask for a PDF summary (
pdf_extract), a site audit (ghost_audit_site) or a code audit (provibe_audit). Each step gets its own receipt. - See the whole session. Ask "Show me everything you did this session."
continuity_get_recentlists the session's receipts in order. - Close the job. Ask the AI to run
workflow_close. That gathers the job's receipts into one summary receipt.
If your AI did something outside Zambo, it can record that with journal_log. Those entries are marked as agent-reported, not as a Zambo-run receipt, so you always know which is which.
What you get
- A verifiable receipt for every Zambo step, with a public link and a verify link.
- One summary receipt per job (
workflow_close) that you can share as a single link. - A handover link when the work is for someone else. Say "Use zambo to hand this over to my client" and the
handovertool packs the receipts into one page with Verify all and Accept / Flag. - An honest boundary: proof that steps happened, not a promise that answers are right.
Related
- Did my agent lie?
- Verify a receipt
- Receipts
- What Zambo receipts verify
- AI agent fake completion
- How to verify AI agent work (developer guide)
- Connect your AI
FAQ
How do I prove my AI actually did a task?
Run the task through Zambo. Each tool call returns a verifiable receipt at a public link, which shows what ran, what came back and when.
Does a receipt prove the AI's answer is correct?
No. It shows the saved output hasn't changed since the call was recorded. You still judge whether the output is right.
Do I need an API key or account?
No. Paste zambo.dev/mcp into your AI. Most tools are free up to 20 calls per tool per day.
Can I see every step my AI took?
Yes, for steps run through Zambo. Ask for the session's recent activity and you get the receipts in order. workflow_close turns them into one summary receipt.
What about work my AI did outside Zambo?
It can log that with journal_log. Those entries are marked as agent-reported, so you can tell them apart from Zambo-run receipts.
CTA: Paste zambo.dev/mcp into your AI. Then ask for one live price and open your first receipt.