diff --git a/doc/PodioInputOutput.md b/doc/PodioInputOutput.md
new file mode 100644
index 00000000..1263d35c
--- /dev/null
+++ b/doc/PodioInputOutput.md
@@ -0,0 +1,110 @@
+# Reading and writing EDM4hep files in Gaudi
+
+The facilities to read and write EDM4hep (or in general event data models based
+on podio) are provided by [`k4FWCore`](https://github.com/key4hep/k4FWCore).
+This page will describe their usage, but not go into too much details of their
+internals. This page also assumes a certain familiarity with Gaudi, i.e. most of
+the snippets just show a minimal configuration part, and not a complete runnable
+example.
+
+## The `k4DataSvc`
+
+Whenever you want to work with EDM4hep in the Gaudi based framework of Key4hep,
+you will need to use the `k4DataSvc` as *EventDataSvc*. You can instantiate and
+configure this service like the following
+
+```python
+from Gaudi.Configuration import *
+from Configurables import k4DataSvc
+
+evtSvc = k4DataSvc("EventDataSvc")
+```
+
+**It is important that the name is `EventDataSvc` in this case, as otherwise
+this is an assumption from Gaudi.** Once you have the `k4DataSvc` instantiated,
+you still have to make the `ApplicationMgr` aware of it, by making sure that the
+`evtSvc` is in the list of the *external services* (`ExtSvc`):
+
+```python
+from Configurables import ApplicationMgr
+ApplicationMgr(
+    # other args
+    ExtSvc = [evtSvc]
+)
+```
+
+## Reading events
+
+To read events you will need to use the `PodioInput` algorithm in addition to
+the [`k4DataSvc`](#the-k4datasvc). Currently, you will need to pass the input
+file to the `k4DataSvc` via the `input` option but pass the collections that you
+want to read to the `PodioInput`. We are working on making this (discussion
+happens in this [issue](https://github.com/key4hep/k4FWCore/issues/105)). The
+parts of your options file related to reading EDM4hep files will look something
+like this
+
+```python
+from Configurables import PodioInput, k4DataSvc
+
+evtSvc = k4DataSvc("EventDataSvc")
+evtSvc.input = "/path/to/your/input-file.root"
+
+podioInput = PodioInput()
+podioInput.collections = [
+    # the complete list of collection names to read
+]
+```
+
+**Note that currently only the collections that are inside the `collections`
+list will be read and become available for later algorithms.**
+
+It is possible to change the input file from the command line via
+```bash
+k4run <your-options-file> --EventDataSvc.input=<input-file>
+```
+
+## Writing events
+
+To write events you will need to use the `PodioOutput` algorithm in addition to
+the [`k4DataSvc`](#the-k4datasvc):
+
+```python
+from Configurables import PodioOutput
+
+podioOutput = PodioOutput("PodioOutput", filename="my_output.root")
+```
+
+By default this will write the complete event contents to the output file.
+
+### Writing only a subset of collections
+
+Sometimes it is desirable to limit the collections to a subset of all available
+collections from the EventStore. The `PodioOutput` allows to do this via the
+`outputCommands` option that takes a list of `keep` or `drop` commands. Each
+command must consist of the `keep`/`drop` command and a target. The target is a
+collection name that may include the `?` or `*` wildcard patterns. This might
+look like the following
+
+```python
+podioOutput.outputCommands = ["keep *"]
+```
+
+which will keep everything (the default), while 
+
+```python
+podioOutput.outputCommands = ["drop *"]
+```
+
+will simply drop all collections and effectively write an empty file (apart from
+some metadata). A common pattern is to `"drop *"` and then selectively adding
+`keep` collections to keep, e.g. to only keep the highest level MC and reco
+information:
+
+```python
+podioOutput.outputCommands = [
+    "drop *",
+    "keep MCParticlesSkimmed",
+    "keep PandoraPFOs",
+    "keep RecoMCTruthLink",
+]
+```