# Echo (/docs/ai/echo)

Echo is a tiny screen in the middle of the page header. It tells what is going on in the window as it happens: a
send and its timing, an error, the Runner's progress, a flow waiting for you, an agent's call, a switch to a
protected environment. Now and then it adds a joke, as much as you like. Click it for a report of the session, jokes
aside. Echo is on by default, and you can turn it off.

<Shot id="ai-echo-screen" alt="Echo in the middle of the page header, showing the status and time of the request just sent" />

## Turn Echo on or off [#turn-echo-on-or-off]

Echo is on by default. Open **Settings › General**, and under **Echo** use **Show Echo's status screen in the page
header**. Every workspace window follows it. When the header is too narrow for the screen beside the page's title and
buttons, Echo steps aside until there is room again.

To hide it again, turn the setting off, or Control-click the screen and choose **Hide Echo**.

## What Echo tells [#what-echo-tells]

Echo watches the window and puts a short line on its screen for:

- **Sends**: the request being sent and how long it has taken so far, then the status and time of the response, the
  checks that failed, or why there was no response (a timeout, nothing listening at the address, a host name that
  didn't resolve, a failed TLS handshake). A cancelled send too.
- **Sessions**: a WebSocket, Socket.IO or MQTT session connecting and closing.
- **The Runner**: progress while it runs, then how many requests passed or failed, or where you stopped it.
- **Flows**: a flow running, waiting for you, and how it ended.
- **Mocks**: a mock server starting and stopping, and the requests it answers.
- **Environments**: a switch of the active environment, and a warning when it is a
  [protected one](/docs/workspaces/protected-environments).
- **Agents**: an agent connecting, its calls, and approvals waiting for you; the Assistant answering.
- **What's New**: after an update, the releases you haven't read yet.

### How much it jokes [#how-much-it-jokes]

Choose with **Humor** in **Settings › General › Echo**, or Control-click the screen:

| Humor | What Echo does |
| --- | --- |
| **Facts only** | Status and details, no jokes. |
| **A little** | A joke now and then. The default. |
| **Plenty** | A joke on every line, and more chatter while nothing happens. |

## The report [#the-report]

Click Echo, or Control-click it and choose **Show Echo's Report**. The **SESSION REPORT** unrolls from the screen with
the facts of the session since it started:

<Shot id="ai-echo-report" alt="Echo's session report: sends, failures, average, slowest and fastest times, runs, and a log of the session's events" />

| Readout | What it counts |
| --- | --- |
| **SENDS**, **FAILED** | Requests sent, and how many failed |
| **AVERAGE**, **SLOWEST**, **FASTEST** | Response times; hover over the slowest or fastest to see the request |
| **RUNS**, **RUN FAILURES** | Runner runs, and failed requests out of all the requests they ran |
| **FLOWS** | Flow runs, and how many failed |
| **SESSIONS** | WebSocket, Socket.IO and MQTT sessions |
| **MOCK CALLS** | Requests your mock servers answered |
| **AGENT CALLS** | Calls from agents |

Readouts for things that didn't happen stay out. Below them is the log of the session's events with their times.
**Clear** starts the report afresh. Close it with the close button or <kbd>Esc</kbd>.

## AI comments [#ai-comments]

Echo writes its own lines. If you like, a model can write some of its comments instead: turn on **Let AI write some
of Echo's comments** in **Settings › General › Echo**. It is off by default, and not available with **Facts only**.

- On a Mac with Apple Intelligence (macOS 26), Apple's on-device model writes them, and nothing leaves the Mac.
- Otherwise the model checked in **Settings › Assistant** writes them. **Written by** shows which. The first time,
  Echolot asks the [one-time permission](/docs/ai/assistant#the-one-time-permission) for that provider. If you
  haven't allowed it, **Written by** says so with an **Allow…** button, and Echo writes its own lines meanwhile. A
  permission you already gave the Assistant covers Echo too. CLI agents never write Echo's comments.

A model writes at most one comment every 30 seconds, and only about failures and other notable moments: a failed or
very slow (3 seconds or more) or very fast (under 30 ms) send, a failed check, a send to or a switch to a protected
environment, a finished run or flow, an agent waiting for approval.

### What the model hears [#what-the-model-hears]

Only what happened, in a sentence: the kind of event, a request's kind and method, a status code, a duration, how
many checks failed, why a send got no response (timeout, nothing listening, host name, TLS), that a run of so many
requests had so many failures. For example: *"A HTTP POST request got status 422 in 180 ms."* or *"A test run of 12 requests finished with
2 failures."*

Never names, addresses, headers, bodies, variables or secrets.

## Accessibility [#accessibility]

- With **Reduce Motion** on (System Settings › Accessibility › Display), lines appear at once instead of typing out
  and scrolling, the screen only redraws when its text changes or a clock ticks, and the report opens with a short
  fade instead of unrolling.
- With **Increase Contrast** on, the screen draws with more contrast.
- VoiceOver reads Echo's current lines; the screen's label is "Echo".