Rework docker setup, split ponos

content/deployment/ponos.md

++++
+title = "Ponos agent"
+description = "Deploy Ponos agents on your own infrastucture to run machine learning processes on GPU"
+
+weight = 50
++++
+
+{% info() %}
+This feature is only available for Enterprise Edition customers.
+{% end %}
+
+If your setup requires Machine Learning process, you'll need at least one **Ponos Agent** on a dedicated server.
+
+The setup of this kind of server is easier, as it only requires to run the agent (from Docker image `registry.gitlab.teklia.com/arkindex/ponos-agent`) and configure it. The tasks will then be triggered by the agent automatically.
+
+To begin the setup, you'll need 2 private keys: one for the backend, another for the agent. Each agent needs a dedicated key to authenticate itself.
+
+To generate a valid private key: 
+
+```sh
+openssl ecparam -name secp384r1 -genkey -noout > agent.key
+```
+
+A YAML configuration file is also required:
+
+```yaml
+---
+# Save as agent.yml
+url: https://ark.localhost/
+farm_id: XXXXX
+seed: YYYYY
+data_dir: /data
+private_key: /etc/ponos/agent.key
+```
+
+The `farm_id` and `seed` information can be found in the Arkindex administration interface under the section **Ponos > Farms**.
+
+You can then run the agent as:
+
+```sh
+docker run \
+  --name=ponos \
+  -v /var/run/docker.sock:/var/run/docker.sock \
+  -v ./agent.yml:/etc/ponos/agent.yml:ro \
+  -v ./agent.key:/etc/ponos/agent.key:ro \
+  -v ponos_data:/data
+  registry.gitlab.teklia.com/arkindex/ponos-agent:X.Y.Z
+```
+
+Please note that the agent requires a write access on the local Docker socket in order to create new containers that will run the tasks.
+
+The `ponos_data` docker volume is not required, but will allow to retrieve debug logs outside of the agent container.

content/deployment/setup.md

 +++
-title = "Docker setup"
+title = "Setup with Docker"
 description = "Deploy Arkindex on your own infrastucture using Linux and docker-compose"
 
 weight = 30
 +++
 
+In this documentation, we'll provide information on how to setup **Arkindex Community Edition** through Docker, on a single server.
+
+If you need more information for a cluster or cloud setup, or using Enterprise Edition, please [contact us](https://teklia.com/contact/).
+
+
 ## Requirements
 
-- A bare metal server running Linux Ubuntu LTS (20.04 or 22.04) for the platform
-   - If you plan to run Machine Learning processes, you'll need another server
+- A bare metal server running Linux Ubuntu LTS (22.04) for the platform
+   - If you plan to run Machine Learning processes, you'll need another server with a GPU
 - [Docker installed on that server](https://docs.docker.com/desktop/install/linux-install/)
-- [docker-compose](https://docs.docker.com/desktop/install/linux-install/)
+- A domain name for that server
 
 ## Third-party services
 
@@ -28,13 +33,19 @@ Optional services:
 - a IIIF server for your images, [cantaloupe](https://cantaloupe-project.github.io/)
 - a search engine to lookup your transcriptions: [Apache Solr](https://solr.apache.org/)
 
+You can find a detailed [docker-compose](https://gitlab.teklia.com/arkindex/backend/-/tree/master/docker?ref_type=heads) file in the Arkindex backend repository.
+
+Of course, you'll need to tweak the file so that it matches your own settings and domain name:
+- replace all `ark.localhost` references with your own domain name
+- remove the minio references if you are using Amazon S3 buckets (or any other compatible solution)
+- remove solr references if you do not need search capabilities
+
 
 ## Arkindex software
 
 Teklia will provide you with several docker images (to load using [docker load](https://docs.docker.com/engine/reference/commandline/load/)):
 - the backend image, tagged `registry.gitlab.teklia.com/arkindex/backend:X.Y.Z`, must be present on your application server,
 - the tasks image, `registry.gitlab.teklia.com/arkindex/tasks:X.Y.Z`, will be used to by the remote workers (file imports, thumbnails generation, ...).
-- the ponos image, `registry.gitlab.teklia.com/arkindex/ponos-agent:X.Y.Z` will be used to actually run the asynchronous tasks across all your remote workers.
 - 
 {{ figure(image="deployment/stack.png", height=250, caption="Arkindex Platform and a single Worker") }}