add caching explanation

Author: dobrac

template/how-it-works.mdx

@@ -25,4 +25,23 @@ We call this sandbox snapshot a _sandbox template_.
 </Note>
 
 ## Caching
-TODO
+The caching concept is similar to [Docker's layer caching](https://docs.docker.com/build/cache/). For each layer command (.copy(), .runCmd(), .setEnvs(), etc.), we create a new layer on top of the existing.
+Each layer is cached based on the command and its inputs (e.g., files copied, command executed, environment variables set).
+If a layer command is unchanged and its inputs are the same as in any previous build, we reuse the cached layer instead of rebuilding it.
+
+This significantly speeds up the build process, especially for large templates with many layers.
+The cache is scoped to the team, so even if you have multiple templates, they can share the same cache if they have identical layers.
+
+### Files Caching
+When using the `.copy()` command, we cache the files based on their content. If the files haven't changed since the last build, we reuse them from the files cache.
+
+We differ from Docker here. Because we build the template on our infrastructure, we use improved caching on files level.
+Even if you invalidate the layer before `.copy()` (e.g., by changing ENV variables), we'll reuse the already uploaded files.
+The `copy()` command will still be re-executed, but the files for the layer will be reused from the files cache, no need to upload them from your computer again.
+
+### Use Case for Caching
+You can leverage caching to create templates with multiple variants (e.g., different RAM or CPU) while reusing the common layers.
+When building the template, just change the template alias to a specific RAM/CPU configuration (e.g., `my-template-2cpu-2gb`, `my-template-1cpu-4gb`), keep the rest of the template definition the same, and the build process will reuse the cached layers.
+
+### Optimize Build Times
+To optimize build times, place frequently changing commands (e.g., copying source code) towards the end of your template definition. This way, earlier layers can be cached and reused more often.