> ## Documentation Index
> Fetch the complete documentation index at: https://docs.condense.io/llms.txt
> Use this file to discover all available pages before exploring further.
# Autoscaling
> Condense adds and removes copies of your custom connector for you, and Insights tells you when the CPU and memory it reserves no longer fit what it uses, so you stop paying for idle capacity and avoid out-of-memory restarts.
Every custom connector you deploy on Condense reserves some CPU and memory. Get that wrong in either direction and it shows up in two places: your invoice and the stability of your pipeline.
* **Reserve too much** and you pay every month for capacity nothing uses. Nodes that could have been switched off stay on.
* **Reserve too little** and your connector slows down under load, or is stopped when it runs out of memory.
Autoscaling in Condense looks after both, in two different ways.
```mermaid theme={null}
flowchart LR
SVC["Your connector"] --> COPIES["How many copies run
Condense adjusts it"]
SVC --> SIZE["How big each copy is
You decide, Insights advises"]
```
**How many copies run is handled for you.** Condense adds copies of your connector, called replicas, when it needs them and removes them when it does not, inside the range you set.
**How big each copy is stays your decision.** The CPU and memory each copy reserves are figures you choose, and Condense never changes them behind your back. Insights watches what your connector really uses. When your figures stop fitting, it shows you a better one, and you can apply it in one click.
The rest of this section is about the second part, because that is where you have decisions to make.
> [!NOTE]
>
> Insights is available for custom connectors. The pages in this section say "connector" for short, where the screens say "service".
## What you get
> [card] Pay for what you use
>
> When a connector reserves far more than it uses, Insights shows the smaller figure and tells you that the change lowers your invoice.
> [card] Fewer surprise restarts
>
> When a connector needs more memory than it reserved, or was stopped for running out of it, Insights shows you the figure that fits.
> [card] One click, not a spreadsheet
>
> Each recommendation puts the figure you have now beside the one it suggests. Accept it and the connector restarts once.
> [card] You stay in control
>
> Nothing changes until you accept. Dismiss a recommendation and it stays dismissed until the recommended figure changes.
## Where to find it
Open your workspace, open the pipeline, and select a custom connector. Its **Insights** tab sits beside Observability, Alerts, Logs and Environment Variables.
The Insights tab of a connector with nothing to change.
That tab is where recommendations appear. It is also where you can see that a connector is being watched, or why nothing is being recommended.
You do not have to open every connector to find out. When a connector has recommendations waiting, its card on the pipeline shows a lightbulb, and its Insights tab shows an Insight Available marker with a count.
The four figures themselves, Requests and Limits for CPU and for RAM, are on the **Resource Configurations** section when you deploy a connector and afterwards. You can change them there yourself at any time.
## What Insights will never do
* Change how much your connector reserves by itself. Every change is one you accepted.
* Restart your connector by itself. A restart only follows something you accepted, and you are warned before you commit.
* Add or remove nodes. How much capacity your environment has is not part of this feature.
* Make up a figure. Where it cannot tell, it says so.
## Next steps
> [card] [How Insights follows your connector](https://docs.condense.io/condense/v3.0.0/autoscaling/how-insights-follows-your-connector.md)
>
> From deployment to a change you have accepted.
> [card] [Choose what your connector reserves](https://docs.condense.io/condense/v3.0.0/autoscaling/choose-what-your-connector-reserves.md)
>
> The four figures, and what goes wrong at each end.
> [card] [What Insights can tell you](https://docs.condense.io/condense/v3.0.0/autoscaling/what-insights-can-tell-you.md)
>
> The five situations it speaks up about.
This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.