Name who each agent works for, cap what it can do, and keep proof of every action.
Refund order 4821 for $18.
Operatorfor atlas
Refunded $18.00. I did not email the customer. Sending messages is blocked for me.
Set them once. They hold in every client, on every call.
@atlas12.8K
atlas-dev4.2KRuns, failures, approvals, and the model behind each agent, on one page.
Your sign-in, framework, and keys stay. Proofable decides what each agent may do before it acts.
| Your stack | Proofable adds |
|---|---|
| Sign-in tools like Okta or Clerk prove who logged in. | What each agent may access, spend, or change after that. |
| Frameworks like LangChain or CrewAI run the agent's work. | A permission check before the agent acts. |
| API keys and secrets let an agent reach a system. | A spend cap, an end date, and revoke at any time. |
| Tracing tools like Langfuse record what happened. | Proof another app can check again. |
Each one scoped to its job, capped, and on record.
Reads the ticket, refunds with the saved Stripe key, and stops before it emails the customer. Every refund leaves a proof.
Set up this agentFiles and labels issues from new alerts and posts a summary to the team. It cannot run commands or change its own permissions.
Set up this agentA scheduled job searches the web, pulls the team notes, and writes the brief. The sources and the run are on record.
Set up this agentSign-in names who logged in. Proofable sets what the agent may do, how much it may spend, and when access ends, and checks that before the action.
Keep your framework. LangChain, CrewAI, and similar tools run the agent's reasoning and tools. Proofable decides what the agent is allowed to do and checks that before it acts.
Which actions the agent may take, which it may never take, how much it may spend, and when access ends. A blocked action always wins over an allowed one, and you can revoke access at any time.
The check fails before the action runs. A missing or expired permission fails closed, and an irreversible action can require a person to approve it first.
Add Proofable to Claude, Cursor, or Codex and sign in once. Add your agent in the project. Its permissions appear there.
Yes. A checked action can leave a proof with the agent, its owner, and the result. Other apps and agents can check it without seeing your private data, and you choose what is shared.
Connect the agent you already run. Its identity, limits, and receipts are ready in every connected client.
npx -y @proofable/sdk setup