> ## Documentation Index
> Fetch the complete documentation index at: https://docs.streamwizard.org/llms.txt
> Use this file to discover all available pages before exploring further.
# The auto switcher
> Watches your connection and switches scenes when it goes bad. Here's exactly how it thinks.
You're walking, you go under a bridge, your bitrate falls off a cliff. Chat
stares at a frozen frame for forty seconds. The auto switcher is what stops
that: it watches your incoming signal and moves your OBS to a fallback scene
when the connection goes bad, then comes back once it's properly stable.
You'll find it in the **Auto Switcher** tab on the Cloud OBS page. The
container has to be running to open it; the scene pickers read the real scene
list from your OBS.
## What it reads
The ingest server measures your incoming stream once a second and hands the
switcher three numbers:
* **Bitrate** in kbps. A floor: below it counts as bad.
* **Ping (RTT)** in milliseconds. A ceiling.
* **Packet loss** as a percentage. A ceiling. This is measured as packets
that were actually given up on, damage your viewers can see, not raw
radio loss that SRT quietly repairs. Cellular connections routinely lose
and recover several percent without anyone noticing, so raw loss was
never a useful signal.
Each second, every metric is scored good or bad against its limit. On plain
RTMP only bitrate arrives, so only bitrate is judged. One more reason to
[stream SRT or SRTLA](/irl/irl-pro#srt-or-srtla).
## How it decides
It never reacts to one bad second. Everything runs on streaks:
* **Falling back**: if any one metric has been bad for its trigger count in
a row, the switcher leaves your live scene.
* **Coming back**: all three metrics must be good for the recover count in
a row. One bad second resets the whole streak.
* **Startup**: a fresh stream is checked before it's allowed on your live
scene. Bad from the first second means it opens on the fallback instead
of showing chat a mess.
* **Offline**: if samples stop arriving entirely for the offline timeout,
the stream counts as gone, not just bad, and you land on your connection
lost scene.
Recovery is deliberately slower than falling back. The ingest holds a 4
second buffer, and during an outage that buffer drains; it needs about its
own depth of good delivery to refill. Switch back faster than that and your
viewers get a stuttering picture. That's also why a stream that flaps
between scenes every ten seconds is worse than one that waits a few extra
seconds to be sure.
## The scenes
Two or three, your choice:
* **Live scene**: your normal scene with the feed in it.
* **Connection lost scene**: your BRB screen.
* **Low bitrate scene** (3-scene setup only): for rough-but-alive, maybe a
cropped camera or a "signal is struggling" banner. On a 2-scene setup,
rough and gone both land on the connection lost scene.
Start with two if you're unsure.
Point the pickers at real scenes and save. The switcher tracks scenes by
their internal id, so renaming a scene in OBS later is fine; deleting one
makes the picker tell you it's gone so you can re-pick.
## Sensitivity presets
Three presets set all the numbers at once. **Balanced** is the default and
right for most streams.
The numbers behind them:
| | Relaxed | Balanced | Fast |
| ------------- | --------- | -------- | -------- |
| Bitrate floor | 300 kbps | 500 kbps | 800 kbps |
| Ping max | 2000 ms | 1500 ms | 1000 ms |
| Loss max | 2% | 1% | 0.5% |
| Switch after | 6s bad | 4s bad | 2s bad |
| Back after | 15s good | 8s good | 5s good |
| Startup check | 8s | 6s | 4s |
| Offline after | 10s quiet | 7s quiet | 5s quiet |
The ping ceilings look high next to other tools, and that's on purpose: the
4 second ingest buffer absorbs latency spikes from tower handovers that
would otherwise trigger a switch on an ordinary walk. The bitrate floors
look low for the same reason: adaptive encoders like IRL Pro ride down to
500-1000 kbps in bad spots and stay watchable, so the floor is a last-resort
signal, not the primary one.
Pick **Relaxed** for mobile in a city, where short dips are normal life.
Pick **Fast** for a connection that's usually stable, where a drop means
something actually broke. Save, and the engine picks the change up within a
second, mid-stream, no restart.
You don't need a laptop to change this. [The deck](/irl/deck) has the same
three presets in its bottom bar, so you can loosen things up the moment you
walk into a packed venue and tighten them again on the way out.
## Advanced thresholds
If no preset fits, flip on **Advanced mode** for the raw table. It starts
from whichever preset you had, so flipping the switch changes nothing by
itself. One sample arrives per second, so every count in the table is really
"this many seconds in a row".
* **Limit**: what counts as bad. Bitrate is a floor, ping and loss are
ceilings.
* **Trigger**: seconds a metric must be bad before leaving your live scene.
* **Recover**: seconds every metric must be good before coming back.
* **Startup**: the check a new stream passes before it's allowed on the
live scene.
* **Offline after**: seconds of complete silence that count as gone. Keep
it above 5; OBS still has about 4 seconds of buffered good video when
stats stop, and cutting away sooner abandons a picture that's still fine.
When you edit, change one row at a time; change all three and you learn
nothing. Some rules of thumb:
* Switches too often? Raise **trigger** seconds before you loosen limits.
* Comes back too eagerly and drops again? Raise **recover**. But don't
park it at 20-30 seconds: the good streak resets to zero on a single bad
sample, so a huge recover count can leave a perfectly usable stream stuck
on the backup scene.
* Keep the bitrate floor well under what you normally stream at, or every
small dip becomes a scene change.
## The extras
Four options under the settings, all optional. Logging is the only one on
by default.
* **Logging** puts every switch in your activity feed and VOD timeline with
the reason. Useful the morning after, when you want to know whether it
was you or the connection.
* **Chat notices** let the StreamWizard bot tell chat what's going on so
you don't have to. You write three messages: quality dropped, signal
lost, back live. Four placeholders work in all of them: `{bitrate}` in
kbps, `{rtt}` in ms, `{loss}` as a percentage, `{scene}` for where it
switched. Chat gets told once per switch, not once per second.
* **Warning source** is a source in your live scene, an "unstable
connection" banner for example, that appears after two bad seconds and
disappears after five good ones, without any scene switch. The polite
warning before the real fallback.
* **Auto stop** ends the OBS stream output after a set number of minutes
offline, 1 to 120. Nobody needs three hours of BRB loop. Your container
and scenes are left alone.
There's also a field to **pin one stream key label**, so the switcher
judges that stream only and ignores anything else arriving at your ingest.
## The status card
While a stream is arriving, the status card shows the state, which stream
key it's watching, and three progress bars, one per metric. Live, the bars
fill toward a switch; recovering, they fill toward going back. The same
status is available as an overlay widget in the Overlay Editor if you want
it on stream. The quickest way in is **New Overlay → Auto switcher monitor**,
which drops the widget on a canvas already set up for it. It hides itself
while everything is healthy, so it costs you nothing on screen until it has
something to say.
A red **Problem** line is the engine's last error, usually a scene switch
that couldn't reach your OBS. It retries every few seconds by itself; if it
sticks, restart the container.
## Watching it from the street
You can't see the status card while you're out filming. The **Auto switcher
monitor** overlay puts the same information on your phone's own preview,
visible to you and not to your viewers. See
[switcher status on your phone](/irl/switcher-monitor).
## When nothing switches
Work down this list:
1. Is the switcher on, and did you actually hit save?
2. Are scenes picked? Without a live and a connection lost scene there's
nothing to switch between.
3. Does the badge say **OBS Connected**?
4. Is a stream arriving, or does the status card say "Waiting for stream"?
5. Are you holding a scene? "Held by you" means it's standing down on
purpose.
6. "This scene no longer exists" means you deleted it in OBS. Pick a new
one and save.
Still stuck? Open a ticket in [our Discord](https://discord.gg/streamwizard).
We answer fast, usually faster than your stream takes to reconnect under a
bridge. Bring your stream key label and roughly when it happened, and we'll
dig through the logs with you.