Pushary
Blog
Product

See exactly what your agents did: receipts, a what-changed view, and exports

Per-session receipts, a what-changed view, exportable history, and a record of which surface answered each agent decision.

AG
Aadil Ghani
Founder, Pushary
Jun 7, 20264 min read
Share

An AI agent ran for 40 minutes while you were away. It edited files, ran shell commands, maybe pushed a branch. When you sit back down, the only record is a wall of terminal scrollback, and half of it has already scrolled off. If someone asks what the agent actually changed, you are reconstructing it from memory.

Pushary now gives every agent session a receipt. Structured tool-action metadata, a what-changed view, exportable history, and a record of which surface answered each decision. The notification gets you back in the loop. The receipt tells you what happened while you were gone.

Key takeaways

  • Each session has a receipt with structured tool-action metadata, not raw scrollback.
  • The what-changed view summarizes the edits, commands, and approvals in one place.
  • Every decision records which surface answered it: phone, web, or Slack.

The problem with terminal scrollback

Terminal output is a stream, not a record. It mixes the agent's reasoning, tool calls, and your own commands into one feed that you cannot query, cannot filter, and cannot hand to anyone else. The moment the buffer fills, the early part of the session is gone.

That is fine when you are watching live. It falls apart the second you step away, which is the entire reason you connected a notification tool in the first place. If an agent can run without you, the record of what it did has to outlive the terminal window.

What a receipt contains

A receipt is the structured version of a session. Instead of a text dump, each tool action is stored as its own entry with metadata: which tool ran, the arguments it ran with, whether it was auto-approved, gated for your approval, or denied, and what you sent back if you denied it with a reason.

Because the data is structured, the what-changed view can roll a session up into something readable. Which files the agent touched. Which shell commands it ran. Which actions hit a permission policy and how each one resolved. You get the shape of the session without scrolling through every line of it.

Session titles come from the agent's own prompt, captured through the UserPromptSubmit hook, so a receipt is labeled with the task you actually asked for instead of a generic session ID. When several agents run at once, each one keeps its own receipt with its own name.

Who answered, and where

Approvals can come from a lot of places now. The phone app, the web decision page, or a Slack message. When a decision matters later, "it was approved" is not enough. You want to know which surface approved it.

Every decision records its answer source. A receipt shows that git push was approved from the phone at the lock screen, or that a file delete was denied from Slack with a reason attached. For a team, that is the difference between an audit trail and a guess.

Export when you need the record elsewhere

History is exportable. When you need the record outside Pushary, for a compliance review, an incident write-up, or your own files, you can take it with you instead of screenshotting a dashboard.

The export carries the structured detail, not a flattened summary. Each tool action, the arguments it ran with, how the policy resolved it, and which surface answered. That is the version you want to hand to a reviewer.

The receipt is built from the same data the permission policy engine uses. The read-only safe-floor that auto-approves proven read-only commands was decided from 1,721 real production questions, and every one of those decisions lands in the audit trail the same way.

Why the receipt is the moat

Native notifications are table stakes now. Most agent tools can ping you when they finish. What they do not give you is a durable, structured, exportable account of what the agent did and how each risky step was decided.

That account is what turns "my agent runs while I am away" into something an engineering lead or a compliance reviewer can actually sign off on. The permission policy decides what an agent may do on its own. The audit trail records what it did. Together they are the reason to run Pushary instead of a bare notification.

One honest caveat. Pushary is GDPR-aligned but self-assessed. There is no SOC 2 or ISO certification behind it. The receipts and exports give you the raw material for your own review process; they are not a substitute for a formal attestation. More on how we handle this on the security page.

How to turn it on

Receipts are on by default once an agent is connected. Connect Claude Code, Codex, Gemini CLI, Cursor, or Hermes through the CLI hook and the audit trail starts filling on the first session. The quickstart walks through the setup, and the audit log docs cover the receipt, the what-changed view, and exports in detail.

If you want to see the full control panel, policy plus audit plus the fleet board, start on the AI agents page. A notification tells you an agent needs you. The receipt is the part you reach for afterward, when someone asks what it actually did.

AG
Aadil Ghani
Founder, Pushary

Building Pushary so an AI agent can reach you on your phone and wait for a yes before it does something you would not want.

Read next

Product

How permission policies learn your defaults

Match rules on what the agent is about to run, not just the tool name, and let your real approve and deny history write the rules for you.

Jun 25, 20264 min readAadil Ghani
Product

The Pushary mobile app: answer approvals and send commands from your phone

The phone is now two-way: approve an agent action from the lock screen, then queue an instruction back to the running session.

Jun 16, 20264 min readAadil Ghani
Product

One board for every running agent

Delivery modes, presence-aware routing, and a Fleet board that groups concurrent agent sessions into lifecycle columns.

Jun 14, 20264 min readAadil Ghani

Get a push the moment your agent needs you

Approvals, done alerts, and a kill switch for Claude Code, Codex, Cursor, and the rest. It takes a couple of minutes to set up.