Installation Guide¶

On-premises, also known as self-hosted, is a setup that allows Konfuzio Server to be implemented 100% on your own infrastructure or any cloud of your choice. In practice, it means that you know where your data is stored, how it’s handled and who gets hold of it.

A common way to operate a production-ready and scalabe Konfuzio installation is via Kubernetens. An alternative deployment option is the Single VM setup via Docker. We recommend to use the option which is more familiar to you. In general

On-Premise Konfuzio installations allow to create Superuser accounts which can access all Documents, Projects and AIs via a dedicated view as well as creating custom Roles

Konfuzio Server has beens successfully installed on various clouds.

Here you find a few examples:

Amazon Web Services (AWS)
Microsoft Azure
Google Cloud Platform (GCP)
IBM Cloud
Oracle Cloud Infrastructure (OCI)
Alibaba Cloud
VMware Cloud
DigitalOcean
Rackspace
Salesforce Cloud
OVH Cloud
Hetzner
Telekom Cloud

Billing and License¶

When you purchase a Konfuzio Server self-hosted license online, we provide you with the necessary credentials to download the Konfuzio Docker Images. If you deploy via Kubernetes you can find the Helm Charts already here. An essential part of this process is the unique BILLING_API_KEY. This key should be set as an environment variable when starting the Konfuzio Server using the following information.

Setup Billing API¶

The BILLING_API_KEY needs to be passed as environment variable to the running Docker container. This is inline with the fact that all configuration of Konfuzio is done via environment variables.

Here is an example command to illustrate the setting of the BILLING_API_KEY environment variable:

docker run -e BILLING_API_KEY=your_api_key_here -d your_docker_image

In the command above, replace your_api_key_here with the actual billing API key and your_docker_image with the name of your Docker image.

Here’s what each part of the command does:

docker run: This is the command to start a new Docker container.
-e: This option allows you to set environment variables. You can also use “–env” instead of “-e”.
BILLING_API_KEY=your_api_key_here: This sets the “BILLING_API_KEY” environment variable to your actual API key.
-d: This option starts the Docker container in detached mode, which means it runs in the background.
your_docker_image: This is the name of your Docker image. You replace this with the actual name of your Docker image.

Instead of using “plain” Docker we recommend to use our Helm Chart or Docker-Compose to install Konfuzio Server.

For Docker-Compose you can simply replace the BILLING_API_KEY placeholder in the docker-compose.yaml file.
When using Helm on Kubernetes, the BILLING_API_KEY is set as ‘envs.BILLING_API_KEY’ in the values.yaml file.

The Konfuzio container continues to report usage to our billing server, i.e., app.konfuzio.com, once a day. We assure you that the containers do not transmit any customer data, such as the image or text that’s being analyzed, to the billing server.

Important Updates¶

If the unique key associated with the Contract on app.konfuzio.com is removed, your self-hosted installation will cease to function within 24 hours. Currently, there is no limit on the number of pages that can be processed within this time. The Konfuzio Server will also refuse to start if any Python file within the container has been modified to preserve the integrity of our license checks.

Technical Background¶

Our server restarts itself after 500 requests by default. The code that prevents changes to Python files in the container has been obfuscated to ensure a significant challenge for anyone attempting to modify the source code.

Air-gapped Environments and Upcoming Changes¶

For those operating the Konfuzio Server in air-gapped environments, your Konfuzio Docker images are licensed to operate for one year, based on the release date, without needing to connect to our billing server.

Soon, however, we will be introducing changes to our license check mechanism. The entire license check will be deactivated if a BILLING_API_KEY is not set. We are currently notifying self-hosted users about this upcoming change, urging them to configure their billing credentials. For more information, please refer to the Billing and License section in our documentation.

For users requiring further instructions, especially those operating in air-gapped environments, please contact us for assistance.

We appreciate your understanding and cooperation as we continue to enhance our security, licensing, and billing measures for all self-hosted installations of the Konfuzio Server.

Kubernetes¶

Required tools¶

Before deploying Konfuzio to your Kubernetes cluster, there are some tools you must have installed locally.

kubectl¶

kubectl is the tool that talks to the Kubernetes API. kubectl 1.15 or higher is required and it needs to be compatible with your cluster (+/-1 minor release from your cluster).

> Install kubectl locally by following the Kubernetes documentation.

The server version of kubectl cannot be obtained until we connect to a cluster. Proceed with setting up Helm.

Helm¶

Helm is the package manager for Kubernetes. Konfuzio is tested and supported with Helm v3.

Getting Helm¶

You can get Helm from the project’s releases page, or follow other options under the official documentation of installing Helm.

Connect to a local Minikube cluster¶

For test purposes you can use minikube as your local cluster. If kubectl cluster-info is not showing minikube as the current cluster, use kubectl config set-cluster minikube to set the active cluster. For clusters in production please visit the Kubernetes Documentation.

Initializing Helm¶

If Helm v3 is being used, there no longer is an init sub command and the command is ready to be used once it is installed. Otherwise please upgrade Helm.

Next steps¶

Once kubectl and Helm are configured, you can continue to configuring your Kubernetes cluster.

Deployment¶

Before running helm install, you need to make some decisions about how you will run Konfuzio. Options can be specified using Helm’s --set option.name=value or --values=my_values.yaml command line option. A complete list of command line options can be found here. This guide will cover required values and common options.

Create a values.yaml file for your Konfuzio configuration. See Helm docs for information on how your values file will override the defaults. Useful default values can be found in the values.yaml in the chart repository.

Selecting configuration options¶

In each section collect the options that will be combined to use with helm install.

Secrets¶

There are some secrets that need to be created (e.g. SSH keys). By default they will be generated automatically.

Networking and DNS¶

By default, Konfuzio relies on Kubernetes Service objects of type: LoadBalancer to expose Konfuzio services using name-based virtual servers configured with Ingress objects. You’ll need to specify a domain which will contain records to resolve the domain to the appropriate IP.

--set ingress.enabled=True
--set ingress.HOST_NAME=konfuzio.example.com

Persistence¶

By default the setup will create Volume Claims with the expectation that a dynamic provisioner will create the underlying Persistent Volumes. If you would like to customize the storageClass or manually create and assign volumes,please review the storage documentation.

Important : After initial installation, making changes to your storage settings requires manually editing Kubernetes objects, so it’s best to plan ahead before installing your production instance of Konfuzio to avoid extra storage migration work.

TLS certificates¶

You should be running Konfuzio using https which requires TLS certificates. To get automated certificates using letesencrypt you need to install cert-manager in your cluster. If you have your own wildcard certificate, you already have cert-manager installed, or you have some other way of obtaining TLS certificates. For the default configuration, you must specify an email address to register your TLS certificates.

Include these options in your Helm install command:

--set letsencrypt.enabled=True
--set letsencrypt.email=me@example.com

PostgreSQL¶

By default this Konfuzio provides an in-cluster PostgreSQL database, for trial purposes only.

Note

Unless you are an expert in managing a PostgreSQL database within a cluster, we do not recommended this configuration for use in production.**

A single, non-resilient Deployment is used

You can read more about setting up your production-readydatabase in the PostgreSQL documentation. As soon you have an external PostgreSQL database ready, Konfuzio can be configured to use it as shown below.

Include these options in your Helm install command:

--set postgresql.install=false
--set global.psql.host=production.postgress.hostname.local
--set global.psql.password.secret=kubernetes_secret_name
--set global.psql.password.key=key_that_contains_postgres_password

Redis¶

All the Redis configuration settings are configured automatically.

Persistent Volume¶

Konfuzio relies on object storage for highly-available persistent data in Kubernetes. By default, Konfuzio uses a persistent volume within the cluster.

CPU, GPU and RAM Resource Requirements¶

The resource requests, and number of replicas for the Konfuzio components in this setup are set by default to be adequate for a small production deployment. This is intended to fit in a cluster with at least 8 vCPU with AVX2 support enabled, 32 GB of RAM and one Nvidia GPU with minimum 4GB which supports at least CUDA10.1 and CUDNN 7.0. If you are trying to deploy a non-production instance, you can reduce the defaults in order to fit into a smaller cluster. Konfuzio can work without a GPU. The GPU is used to train and run Categorization AIs. We observe a 5x faster training and a 2x faster execution on GPU compared to CPU.

Storage Requirements¶

This section outlines the initial storage requirements for the on-premises installation. It is important to take these requirements into consideration when setting up your server, as the amount of storage needed may depend on the number of documents being processed.

For testing purposes, a minimum of 10 GB is required per server (not per instance of a worker).
For serious use, a minimum of 100 GB should be directly available to the application. This amount should also cover the following:
- Postgres, which typically uses 10% of this size.
- Docker image storage, up to 25 GB should be reserved for upgrades.
Each page thumbnail adds 1-2 KB to the file size.
After uploading, the total file size of a page image and its thumbnails increases by approximately a factor of 3 (10 MB becomes approximately 30 MB on the server).
To reduce storage usage, it is recommended to disable sandwich file generation by setting ALWAYS_GENERATE_SANDWICH_PDF=False.

Deploy using Helm¶

Once you have all of your configuration options collected, we can get any dependencies and run Helm. In this example, we’ve named our Helm release konfuzio.

helm repo add konfuzio-repo https://git.konfuzio.com/api/v4/projects/106/packages/helm/stable
helm repo update
helm upgrade --install konfuzio konfuzio-repo/konfuzio-chart --values my_values.yaml

Please create a my_values.yaml file for your Konfuzio configuration. Useful default values can be found in the values.yaml in the chart repository. See Helm docs for information on how your values file will override the defaults. Alternativ you can specify you configuration using --set option.name=value.

Monitoring the Deployment¶

The status of the deployment can be checked by running helm status konfuzio which can also be done while the deployment is taking place if you run the command in another terminal.

Autoscaling¶

The Konfuzio Server deployments can be scaled dynamically using a Horizontal Pod Autoscaler and a Cluster Autoscaler. The autoscaling configuration for the Konfuzio Server installation of https://app.konfuzio.com can be found in this Helm Chart.

Quick Start via Kubernetes and Helm¶

The following commands allow you to get a Konfuzio Server installation running with minimal configuration effort and relying on the default values of the Chart. This uses Postgres, Redis and S3 via MinIO as in-cluster deployments. This setup is not suited for production and may use insecure defaults.

helm repo add konfuzio-repo https://git.konfuzio.com/api/v4/projects/106/packages/helm/stable
helm repo update
helm install my-konfuzio konfuzio-repo/konfuzio-chart
  --set envs.HOST_NAME="host-name-for-you-installation.com"  \
  --set envs.BILLING_API_KEY="******"
  --set image.tag="released-******"
  --set image.imageCredentials.username="******"  \
  --set image.imageCredentials.password="******"

Upgrade¶

Before upgrading your Konfuzio installation, you need to check the changelog corresponding to the specific release you want to upgrade to and look for any that might pertain to the new version.

We also recommend that you take a backup first.

Upgrade Konfuzio following our standard procedure,with the following additions of:

Check the change log for the specific version you would like to upgrade to
Ensure that you have created a PostgreSQL backup in the previous step. Without a backup, Konfuzio data might be lost if the upgrade fails. 3a. If you use a values.yaml, update image.tag=”released-**” to the desired Konfuzio Server version.
```
helm install --upgrade my-konfuzio konfuzio-repo/konfuzio-chart -f values.yaml
```

3b. If you use “–set”, you can directly set the desired Konfuzio Server version.

helm install --upgrade my-konfuzio konfuzio-repo/konfuzio-chart --reuse-values --set image.tag="released-******"

We will perform the migrations for the Database for PostgreSQL automatically.

Docker - Single VM setup¶

Konfuzio can be configured to run on a single virtual machine, without relying on Kubernetes. In this scenario, all necessary containers are started manually or with a container orchestration tool of your choice.

VM Requirements¶

We recommend a virtual machine with a minimum of 8 vCPU (incl. AVX2 support) and 32 GB of RAM and an installed Docker runtime. A Nvidia GPU is recommended but not required. In this setup Konfuzio is running in the context of the Docker executor, therefore there are no strict requirements for the VMs operating systems. However, we recommend a Linux VM with Debian, Ubuntu, CentOS,or Redhat Linux.

Login to the Konfuzio Docker Image Registry¶

We will provide you with the credentials to login into our Docker Image Registry, which allows you to access and download our Docker Images. This action requires an internet connection.

The internet connection can be turned off once the download is complete. In case there is no internet connection available during setup, the container must be transferred with an alternative method as a file to the virtual machine.

Registry URL: {PROVIDED_BY_KONFUZIO}
Username: {PROVIDED_BY_KONFUZIO}
Password: {PROVIDED_BY_KONFUZIO}

docker login REGISTRY_URL

The Tag “latest” should be replaced with an actual version. A list of available tags can be found here: https://dev.konfuzio.com/web/changelog_app.html.

Quick Start via Docker-Compose¶

The fastest method for deploying Konfuzio Server using Docker is through Docker-Compose. To get started, follow these steps:

Install Docker-Compose in a version that is compatible with Compose file format 3.9.
Run docker compose version to verify that Docker-Compose is available.
Download the docker-compose.yml file.
Fill in the mandatory variables in the first section of the docker-compose file.
Launch Konfuzio Server by running docker compose up -d.

If you prefer to use Docker exclusively, we provide detailed instructions for setting up the containers in each necessary step (Step 1-9).

1. Download Docker Image¶

After you have connected to the Registry, the Konfuzio docker image can be downloaded via “docker pull”.

docker pull REGISTRY_URL/konfuzio/text-annotation/master:latest

The Tag “latest” should be replaced with an actual version. A list of available tags can be found here: https://dev.konfuzio.com/web/changelog_app.html.

2. Setup PostgreSQL, Redis, BlobStorage/FileSystemStorage¶

The databases and credentials to access them are created in this step. Please choose your selected databases at this point. You need to be able to connect to your PostgreSQL via psql and to your Redis server via redis-cli before continuing the installation.

In case you use FileSystemStorage and Docker volume mounts, you need to make sure the volume can be accessed by the konfuzio docker user (uid=999). You might want to run chown 999:999 -R /konfuzio-vm/text-annotation/data on the host VM.

The PostgreSQL database connection can be verified via psql and a connection string in the following format. This connection string is later set as DATABASE_URL.

psql -H postgres://USER:PASSWORD@HOST:PORT/NAME

The Redis connection can be verified using redis-cli and a connection string. To use the connection string for BROKER_URL, RESULT_BACKEND and DEFENDER_REDIS_URL you need to append the database selector (e.g. redis://default:PASSWORD@HOST:PORT/0)

redis-cli -u redis://default:PASSWORD@HOST:PORT

3. Setup the environment variable file¶

Copy the /code/.env.example file from the container and adapt it to your settings. The .env file can be saved anywhere on the host VM. In this example we use “/konfuzio-vm/text-annotation.env”.

4. Init the database, create first superuser via cli and prefill e-mail templates¶

In this example we store the files on the host VM and mount the directory “/konfuzio-vm/text-annotation/data” into the container. In the first step we create a container with a shell to then start the initialization scripts within the container. The container needs to be able to access IP addresses and hostnames used in the .env. This can be ensured using –add.host. In the example we make the host IP 10.0.0.1 available.

docker run -it --add-host=10.0.0.1 \
  --env-file /konfuzio-vm/text-annotation.env \
  --mount type=bind,source=/konfuzio-vm/text-annotation/data,target=/data \
  REGISTRY_URL/konfuzio/text-annotation/master:latest bash

python manage.py migrate
python manage.py createsuperuser
python manage.py init_email_templates
python manage.py init_user_permissions

After completing these steps you can exit and remove the container.

Note

The username used during the createsuperuser dialog must have the format of a valid e-mail in order to be able to login later.

5. Start the container¶

In this example we start four containers. The first one to serve the Konfuzio web application.

docker run -p 80:8000 --name web -d --add-host=host:10.0.0.1 \
  --env-file /konfuzio-vm/text-annotation.env \
  --mount type=bind,source=/konfuzio-vm/text-annotation/data,target=/data \
  REGISTRY_URL/konfuzio/text-annotation/master:latest

The second and third are used to process tasks in the background without blocking the web application. Depending on our load scenario, you might to start a large number of worker containers.

docker run --name worker1 -d --add-host=host:10.0.0.1
  --env-file /konfuzio-vm/text-annotation.env
  --mount type=bind,source=/konfuzio-vm/text-annotation/data,target=/data
  REGISTRY_URL/konfuzio/text-annotation/master:latest
  celery -A app worker -l INFO --concurrency 1 -Q celery,priority_ocr,ocr,
  priority_extract,extract,processing,priority_local_ocr,local_ocr,\
  training,finalize,training_heavy,categorize

docker run --name worker2 -d --add-host=host:10.0.0.1 \
  --env-file /konfuzio-vm/text-annotation.env
  --mount type=bind,source=/konfuzio-vm/text-annotation/data,target=/data \
  REGISTRY_URL/konfuzio/text-annotation/master:latest \
  celery -A app worker -l INFO --concurrency 1 -Q celery,priority_ocr,ocr,\
  priority_extract,extract,processing,priority_local_ocr,local_ocr,\
  training,finalize,training_heavy,categorize

The fourth container is a Beats-Worker that takes care of sceduled tasks (e.g. auto-deleted documents).

docker run --name beats -d --add-host=host:10.0.0.1
  --env-file /konfuzio-vm/text-annotation.env
  --mount type=bind,source=/konfuzio-vm/text-annotation/data,target=/data
  REGISTRY_URL/konfuzio/text-annotation/master:latest
  celery -A app beat -l INFO -s /tmp/celerybeat-schedule

[Optional] 6. Use Flower to monitor tasks¶

Flower can be used a task monitoring tool. Flower will be only accessible for Konfuzio superusers and is part of the Konfuzio Server Docker Image.

docker run --name flower -d --add-host=host:10.0.0.1
  --env-file /konfuzio-vm/text-annotation.env
  --mount type=bind,source=/konfuzio-vm/text-annotation/data,target=/data \
  REGISTRY_URL/konfuzio/text-annotation/master:latest
  celery -A app flower --url_prefix=flower --address=0.0.0.0 --port=5555

The Konfuzio Server application functions as a reverse proxy and serves the Flower application. In order for Django to correctly access the Flower application, it requires knowledge of the Flower URL. Specifically, the FLOWER_URL should be set to http://host:5555/flower.

FLOWER_URL=http://host:5555/flower

graph LR subgraph Network a("User") end subgraph Local Network / Cluster Network a("User") --> e("Konfuzio Server") e("Konfuzio Server") -- FLOWER_URL --> f("Flower") end

Please ensure that the Flower container is not exposed externally, as it does not handle authentication and authorization itself.

[Optional] 7. Run Container for Email Integration¶

The ability to upload documents via email can be achieved by starting a dedicated container with the respective environment variables.

SCAN_EMAIL_HOST = imap.example.com
SCAN_EMAIL_HOST_USER = user@example.com
SCAN_EMAIL_RECIPIENT = automation@example.com
SCAN_EMAIL_HOST_PASSWORD = xxxxxxxxxxxxxxxxxx

docker run --name flower -d --add-host=host:10.0.0.1
  --env-file /konfuzio-vm/text-annotation.env
  --mount type=bind,source=/konfuzio-vm/text-annotation/data,target=/data \
  REGISTRY_URL/konfuzio/text-annotation/master:latest
  python manage.py scan_email

[Optional] 8. Use Azure Read API (On-Premises or as Service)¶

The Konfuzio Server can work together with the Azure Read API. There are two options to use the Azure Read API in an on-premises setup.

Use the Azure Read API as a service from the public Azure cloud.
Install the Azure Read API container directly on your on-premises infrastructure via Docker.

The Azure Read API is in both cases connected to the Konfuzio Server via the following environment variables.

AZURE_OCR_KEY=123456789 # The Azure OCR API key
AZURE_OCR_BASE_URL=http://host:5000 # The URL of the READ API
AZURE_OCR_VERSION=v3.2 # The version of the READ API

For the first option, login into the Azure Portal and create a Computer Vision resource under the Cognitive Services section. After the resource is created the AZURE_OCR_KEY and AZURE_OCR_BASE_URL is displayed. Those need to be added as environment variable.

For the “on-premises” option, please refer to the Azure Read API Container installation guide. Please fill in the Microsoft Request Form to get an API_KEY and ENDPOINT_URL which is compatible with running the On-Premises container. The ENDPOINT_URL must be reachable by the Azure Read API Container.

How to fix common Azure Read API issues for On-Premises¶

1. Ensure AVX2 Compatibility:¶

Verify if your system supports Advanced Vector Extensions 2 (AVX2). You can do this by running the following command on Linux hosts:

grep -q avx2 /proc/cpuinfo && echo AVX2 supported || echo No AVX2 support detected

For further information, refer to Microsoft’s AVX2 support guide.

2. Verify Azure Read Container Status:¶

You can monitor the Azure Read Container’s status through its web interface. To do this, access the container on Port 5000 (i.e. http://localhost:5000/status) from a web browser connected to the Container’s network. You might need to replace ‘localhost’ with the IP/network name of the Azure container. This interface will also help detect issues such as invalid credentials or an inaccessible license server (possibly due to a firewall). Refer to Microsoft’s guide on validating container status for further assistance.

3. Consider Trying a Different Container Tag:¶

If your Azure Read API Container ceases to function after a restart, it may be due to an automatic upgrade to a new, potentially faulty Docker tag. In such cases, consider switching to a previous Docker tag. You can view the complete version history of Docker tags at the Microsoft Artifact Registry. Remember to uncheck the “Supported Tags Only” box to access all available versions.

[Optional] 9. Install the Document Layout Analysis (Segmentation) Container¶

Download the container with the credentials provided by Konfuzio.

Registry URL: {PROVIDED_BY_KONFUZIO}
Username: {PROVIDED_BY_KONFUZIO}
Password: {PROVIDED_BY_KONFUZIO}

docker login REGISTRY_URL
docker pull REGISTRY_URL/konfuzio/text-annotation/detectron2/main:released-2023-08-03_11-39-42
docker run --env-file /path_to_env_file.env REGISTRY_URL/konfuzio/text-annotation/detectron2/main:released-2023-08-03_11-39-42 ./run.sh

The “Document Layout Analysis Container” needs to be started with the following environment variables which you can enter into your .env file

BROKER_URL=  # Set this to an unused Redis database (e.g. "redis://redis:6379/2")
RESULT_BACKEND=  # Set this to an unused Redis database (e.g. "redis://redis:6379/3")

We recommend to run the “Document Layout Analysis Container” with 8GB of RAM. The container can be started and used with less RAM, however this may not work on large images. After the “Document Layout Analysis Container” is running you need to set the DETECTRON_URL for Konfuzio Server to point to the “Document Layout Analysis Container”.

To learn more about how the Document Layout Analysis Container works internally, please click here.

[Optional] 10. Install document summarization container¶

Download the container with the credentials provided by Konfuzio

Registry URL: {PROVIDED_BY_KONFUZIO}
Username: {PROVIDED_BY_KONFUZIO}
Password: {PROVIDED_BY_KONFUZIO}

docker login REGISTRY_URL
docker pull REGISTRY_URL/konfuzio/text-annotation/summarization/main:released-2022-11-29_21-55-57
docker run --env-file /path_to_env_file.env REGISTRY_URL/konfuzio/text-annotation/summarization/main:released-2022-11-29_21-55-57

The summarization container needs to be started with the following environment variables which you can enter into your .env file

GPU=True  # If GPU is present
TASK_ALWAYS_EAGER=False
C_FORCE_ROOT=True
BROKER_URL=  # Set this to an unused Redis database (e.g. "redis://redis:6379/4")
RESULT_BACKEND=  # Set this to an unused Redis database (e.g. "redis://redis:6379/5")

We recommend to run the segmentation container with 4GB of RAM. The segmentation container can be started and used with less RAM, however this may not work on large images. After the summarization container is running you need to set the SUMMARIZATION_URL to point to the segmentation container.

11a. Upgrade to newer Konfuzio Version¶

Konfuzio upgrades are performed by replacing the Docker Tag to the desired version After starting the new containers, database migrations need to be applied by python manage.py migrate and new email-templates need to be initialized python manage.py init_email_templates (see 4.). In case additional migration steps are needed, they will be mentioned in the release notes.

11b. Downgrade to older Konfuzio Version¶

Konfuzio downgrades are performed by creating a fresh Konfuzio installation in which existing Projects can be imported. The following steps need to be undertaken:

Export the Projects that you want to have available after downgrade using konfuzio_sdk. Please make sure you use a SDK version that is compatible with the Konfuzio Server version you want to migrate to.
Create a new Postgres Database and a new Folder/Bucket for file storage which will be used for the downgraded version
Install the desired Konfuzio Server version by starting with 1.)
Import the projects using “python manage.py project_import”

Load Scenario for Single VM with 32GB¶

Scenario 1: With self-hosted OCR¶

Number of Container	Container Type	RAM	Capacity
1	Web Container	4GB	…
3	Generic Celery Worker	4GB	1500 (3 x 500) Pages of Extraction, Categorization, or Splitting per hour
1	Self-Hosted OCR worker	8GB	1200 (1 y 1200) Pages) / hours (Not needed if external API Service is used)
N	remaining Containers	4GB	…

With this setup, around 1200 Pages per hour can be processed using OCR and Extraction, around 750 Pages per hour can be processed if OCR, Categorization and Extraction are active, around 850 Pages per hour can be processed if OCR, Splitting, Categorization and Extraction are active.

Scenario 2: Without self-hosted OCR¶

Number of Container	Container Type	RAM	Capacity
1	Web Container	4GB	…
5	Generic Celery Worker	4GB	2500 (3 x 500) Pages of Extraction, Categorization, or Splitting per hour
N	remaining Containers	4GB	…

With this setup, around 2500 Pages per hour can be processed using OCR and Extraction, around 1250 Pages per hour can be processed if OCR, Categorization and Extraction are active, around 850 Pages per hour can be processed if OCR, Splitting, Categorization and Extraction are active.

Note

In case you train large AI Models (>100 Training Pages) more than 4GB for Generic Celery Workers are needed. The Benchmark used an Extraction AI with “word” detection mode and 10 Labels in 1 Label Set. The capacity is shared between all Users using the Konfuzio Server setup.

The level of parallelity of task processing and therefore throughput can be increased by the number of running “Generic Celery Workers”.

The performance of any system that processes documents can be affected by various factors, including the number of simultaneous connections, the size of the documents being processed, and the available resources such as RAM and processing power.

The number of simultaneous connections can also impact performance. If multiple users are uploading and processing large documents simultaneously, it can put a strain on the system’s resources, leading to slower processing times.

Other factors that can affect performance include network latency, disk speed, and the complexity of the processing algorithms. It’s essential to consider all these factors when designing and deploying document processing systems to ensure optimal performance.

Document Layout Analysis¶

The Konfuzio Server leverages Document Layout Analysis’ capabilities to enhance its Natural Language Processing and Object Detection services. You can read more about it on our blog. The platform’s architecture features multiple Document Layout Analysis containers, each encapsulating a Fast API instance and a Worker. The modular structure of these containers allows scalability based on specific requirements.

Each Fast API instance and its corresponding Worker communicate with a shared Redis instance, a robust in-memory data structure store used for caching and message brokerage. As a high-performance web framework, Fast API is pivotal in constructing efficient APIs.

The process unfolds as follows: The Konfuzio Server forwards document layout analysis tasks to a Container via the Fast API instance. In response, the Fast API generates a task entry in Redis, which a Worker can subsequently pick up and process. Upon completion, the worker returns the results back to the Fast API via Redis, creating a seamless flow of information.

This architecture thus delivers a scalable, high-performance environment equipped to manage substantial loads. It ensures efficient object detection and natural language processing capabilities based on the detectron library, making it an optimal choice for handling complex computational tasks.

graph l("Konfuzio Server") <--> a m("Other Applications") <--> a a("Load-Balancer (Optional)") <--> b a <--> d a <--> f b <--> t("Redis") c <--> t("Redis") d("Redis") <--> t("Redis") e("Redis") <--> t("Redis") f("Redis") <--> t("Redis") g("Redis") <--> t("Redis") subgraph all1["Detectron Container - 1"] b("Fast API") c("Detectron Worker") end subgraph all2["Detectron Container - 2"] d("Fast API") e("Detectron Worker") end subgraph all3["Detectron Container - N"] f("Fast API") g("Detectron Worker") end

Document Layout Analysis is available on https://app.konfuzio.com as well as for self-hosted environment via Docker or on Kubernetes via Helm.

Docker-Compose vs. Kubernetes¶

When it comes to running the Konfuzio Server, the choice between Docker Compose and Kubernetes will depend on your specific requirements and use case.

Docker Compose can be a good choice if you are running Konfuzio Server on a single host in production or for testing and development purposes. With Docker Compose, you can define and customize the services required for the Konfuzio Server in a YAML file we provide, and then use a single command to start all the containers.

On the other hand, Kubernetes is more suitable for production environments where you need to run Konfuzio Server at scale, across multiple hosts, and with features such as auto-scaling and self-healing. Kubernetes has a steep learning curve, but it provides advanced features for managing and scaling containerized applications.

Regarding the use of Docker Compose in multiple VMs, while it’s possible to use Docker Compose in a distributed environment, managing multiple VMs can be more work than using a dedicated orchestration platform like Kubernetes. Kubernetes provides built-in support for managing distributed systems and is designed to handle the complexities of running containers at scale.

In either case, you can use the same Docker image for the Konfuzio Server, which will ensure consistency and portability across different environments. Overall, the choice between Docker Compose and Kubernetes will depend on your specific needs, level of expertise, and infrastructure requirements.

Custom AI model training via CI pipelines¶

Konfuzio is a platform that provides users with the ability to run custom AI workflows securely through its Continuous Integration (CI) pipelines. These pipelines enable users to run their code automatically and continuously, learning from human feedback on any errors or issues as soon as they occur.

In situations where the Kubernetes deployment option is not utilized, Konfuzio recommends using a dedicated virtual machine to run these pipelines. This ensures that the pipelines are isolated from other processes and are therefore less susceptible to interference or interruption.

To effectively run these CI pipelines, the selected CI application must support Docker and webhooks. Docker allows for the creation and management of containers, which are used to package and distribute the code to be tested. Webhooks, on the other hand, provide a way for the CI application to trigger the pipeline automatically.

Furthermore, the CI application needs to have network access to the Konfuzio installation. This is necessary to enable the CI application to interact with the Konfuzio platform, ensuring that the CI pipelines are able to access the necessary resources and dependencies.

Overall, Konfuzio’s use of CI pipelines provides a powerful tool for advanced users to run their AI workflows securely, with the added benefit of automated testing and continuous feedback.

SSO via OpenID Connect (OIDC)¶

Konfuzio utilizes OpenID Connect (OIDC) for identity verification, implementing this through Mozilla’s Django OIDC. OIDC is a well-established layer built atop the OAuth 2.0 protocol, enabling client applications to confirm the identity of end users. Numerous Identity Providers (IdP) exist that support the OIDC protocol, thus enabling the implementation of Single Sign-On (SSO) within your application. A commonly chosen provider is Keycloak. In the subsequent section, you will find a tutorial on how to use Keycloak with OIDC.

SSO Keycloak Integration¶

Keycloak allows single sign-on funtionality. By doing so no user management is done wihtin Konfuzio Server. If you already operate a keycloak server, you can reuse keycloak users.

Set up¶

To start and set up keycloak server:

Download keycloak server
Install and start keycloak server using instruction
Open keycloak dashboard in browser (locally it’s http://0.0.0.0:8080/).
Create admin user
Login to Administration Console
You can add new Realm or use default (Master)

Create new client from Clients navbar item

Fill client form correctly (Access Type and Valid Redirect URIs fields)

Move to Credentials tab and save Secret value
In Users navbar item create users

Environment Variables¶

To integrate konfuzio with keycloak you need to set the following environment variables for you Konfuzio Server installation:

KEYCLOAK_URL (http://127.0.0.1:8080/ - for localhost)
OIDC_RP_SIGN_ALGO (RS256 - by default)
OIDC_RP_CLIENT_ID (client name from 7th point of previous section)
OIDC_RP_CLIENT_SECRET (Secret value from 9th point of previous section)
SSO_ENABLED (set True to activate integration)

Click SSO on login page to log in to Konfuzio using keycloak .. image:: ./keycloak/sso-button.png

Important notes¶

The Keycloak admin user cannot login into Konfuzio Server.

SSO via other Identity Management software¶

In order to connect your Identity Management software with Konfuzio using Single Sign-On (SSO), you’ll need to follow these steps. Please note, this tutorial is meant for Identity Management software that supports OIDC (OpenID Connect), a widely adopted standard for user authentication that’s compatible with many software platforms. If you’re using Keycloak, please refer to the separate tutorial on connecting Konfuzio with Keycloak.

Step 1: Setting up your Identity Provider¶

This step varies depending on the Identity Management software you’re using. You should refer to your Identity Management software’s documentation for setting up OIDC (OpenID Connect). During the setup, you will create a new OIDC client for Konfuzio. You will typically need to provide a client name, redirect URI, and optionally a logo.

The redirect URI should be the URL of your Konfuzio instance followed by “/oidc/callback/”, for example https:///oidc/callback/.

At the end of this step, you should have the following information:

Client ID
Client Secret
OIDC Issuer URL

Step 2: Configuring mozilla-django-oidc¶

You’ll need to set the following environment variables:

OIDC_RP_CLIENT_ID: This should be the client ID from your Identity Management software.
OIDC_RP_CLIENT_SECRET: This should be the client secret from your Identity Management software.
OIDC_OP_AUTHORIZATION_ENDPOINT: This should be your OIDC issuer URL followed by “/authorize”.
OIDC_OP_TOKEN_ENDPOINT: This should be your OIDC issuer URL followed by “/token”.
OIDC_OP_USER_ENDPOINT: This should be your OIDC issuer URL followed by “/userinfo”.
OIDC_OP_JWKS_ENDPOINT: This should be your OIDC issuer URL followed by “/.well-known/jwks.json”.
OIDC_RP_SIGN_ALGO: This is the signing algorithm your Identity Management software uses. This is typically “RS256”.
OIDC_RP_SCOPES: These are the scopes to request. For basic authentication, this can just be “openid email”.

Step 3: Restarting Konfuzio¶

After you have set these environment variables, you should restart your Konfuzio server so that the new settings can take effect. How you do this depends on how you have deployed Konfuzio, but it might involve restarting a Docker container or a Django server.

Step 4: Testing¶

Now you can go to your Konfuzio instance and you should be redirected to your Identity Management software login screen.

Migrate AIs and Projects¶

Overview of Migration Methods¶

This table shows the two migration methods, “Project Export” and “AI File,” for moving various elements Projects and Konfuzio Server environments. The table includes a list of elements such as Annotations, Categories, Label Sets, Labels, and more, and indicates which method is applicable for each element. The “Project Export” method is used for exporting whole Projects, while the “AI File” method is used for exporting elements that are specifically relevant for just running the AI on a different Konfuzio Server environments.

	Project Export	AI File
Annotations	Yes	No
Categories	Yes	No
Categorization AI	No	Yes
All Documents	No	No
Test-/Training Documents	Yes	No
Extraction AI	No	Yes
Label Sets	Yes	No
Labels	Yes	No
Members	No	No

Migrate an Extraction or Categorization AI¶

Superusers can migrate Extraction and Categorization AIs via the webinterface. This is explained on https://help.konfuzio.com.

Migrate a Project¶

Export the Project data from the source Konfuzio Server system.

pip install konfuzio_sdk
# The "init" command will ask you to enter a username, password and hostname to connect to Konfuzio Server.
# Please enter the hostname without trailing backslash. For example "https://app.konfuzio.com" instead of "https://app.konfuzio.com/".
# The init commands creates a .env in the current directory. To reset the SDK connection, you can delete this file.
konfuzio_sdk init
konfuzio_sdk export_project <PROJECT_ID>

The export will be saved in a folder with the name data_. This folder needs to be transferred to the target system The first argument is the path to the export folder, the second is the project name of the imported project on the target system.

python manage.py project_import "/konfuzio-target-system/data_123/" "NewProjectName"

Alternatively, you can merge the Project export into an existing Project.

python manage.py project_import "/konfuzio-target-system/data_123/" --merge_project_id <EXISTING_PROJECT_ID>

Database and Storage¶

Overview¶

To run Konfuzio Server, three types of storages are required. First, a PostgreSQL database is needed to store structured application data. Secondly, a storage for Blob needs to be present. Thirdly, a Redis database that manages the background Task of Konfuzio Server is needed. You can choose your preferred deployment option for each storage type and connect Konfuzio via environment variables to the respective storages. We recommend planning your storage choices before starting with the actual Konfuzio installation.

Storage Name	Recommended Version	Supported Version	Deployment Options
Postgres	Latest Stable	PostgreSQL 12 and higher	Managed (Cloud) Service, VM Installation, Docker, In-Cluster*
Redis	Latest Stable	Redis 6 and higher	Managed (Cloud) Service, VM Installation, Docker, In-Cluster*
Blob Storage	Latest Stable	All with activ support	Filesystem, S3-compatible Service (e.g. Amazon S3, Azure Blob Storage), In-Cluster* S3 via MinIO

*If you use Kubernetes Deployment you can choose the ‘in-Cluster’ option for Postgres, Redis and S3-Storage.

Usage of PostgreSQL¶

Konfuzio Server will create a total of 43 tables and use the following data types. This information refers to release 2022-10-28-07-23-39.

data_type	count
bigint	6
boolean	35
character varying	78
date	2
double precision	29
inet	2
integer	138
jsonb	28
smallint	2
text	21
timestamp with time zone	56
uuid	1

Environment Variables¶

Environment Variables for Konfuzio Server¶

1. General Settings¶

HOST_NAME¶

The HOSTNAME variable is used in the E-Mail templates: https://example.konfuzio.com or http://localhost:8000 for local development.

Note: Please include the protocol (e.g. http://) even the variable is named HOST_NAME.

This is mandatory. Type: string

ALLOWED_HOSTS¶

Default: HOST_NAME variable without the protocol.

A list of strings representing the host/domain names that Konfuzio can serve. This is a security measure to prevent HTTP Host header attacks, which are possible even under many seemingly-safe web server configurations. See https://docs.djangoproject.com/en/4.2/ref/settings/#allowed-hosts

Type: List[string]

CSV_EXPORT_ROW_LIMIT¶

Default: 100

The (approximate) number of rows that an exported CSV file of extraction results can include. Set to 0 for no limit, but beware of timeouts as the CSV files are exported synchronously.

Type: integer

ACCOUNT_BLOCK_FREE_EMAIL_PROVIDERS¶

Default: False

Block registration for users with an email address from a free provider according to the blocklist.

Type: boolean

DETECTRON_URL¶

Default: None

This is used to connect to the optional Document Layout Analysis Container. This is a URL in the form of ‘http://detectron-service:8181/predict’. You might need to adjust the detectron-service to your service name or IP.

Type: string

SUMMARIZATION_URL¶

Default: None

This is used to connect to the optional summarization container. This is a URL in the form of ‘http://summarization-service:8181/predict’. You might need to adjust the summarization-service to your service name or IP.

Type: string

2. Background Tasks via Celery¶

Settings related to Background Tasks.

BROKER_URL¶

Enter a Redis Database. See https://docs.celeryq.dev/en/stable/userguide/configuration.html#std-setting-broker_url

This is mandatory. Type: string

AZURE_OCR_KEY¶

Default: None

Get the API KEY from portal.azure.com for Microsoft.CognitiveServices.

Type: string

AZURE_OCR_BASE_URL¶

Default: None

Get the API URL from portal.azure.com for Microsoft.CognitiveServices

Type: string

AZURE_OCR_VERSION¶

Default: ‘v3.2’

Get the AZURE_OCR_VERSION from portal.azure.com for Microsoft.Cognitives.

Type: string

AZURE_OCR_MODEL_VERSION¶

Default=’latest’

Get the AZURE_OCR_MODEL_VERSION from portal.azure.com for Microsoft.Cognitives.

Type: string

4. Blob Storage Settings¶

DEFAULT_FILE_STORAGE¶

Default: ‘django.core.files.storage.FileSystemStorage’

By default, the file system is used as file storage. To use an Azure Storage Account set ‘storage.MultiAzureStorage’. To use an S3-compatible storage set: ‘storage.MultiS3Boto3Storage’. See https://docs.djangoproject.com/en/4.0/ref/settings/#default-file-storage

Type: string

AZURE_ACCOUNT_KEY:¶

Default: ‘’

The Azure account key.

_This is mandatory if DEFAULT_FILESTORAGE=’storage.MultiAzureStorage’. Type: string

AZURE_ACCOUNT_NAME:¶

Default: ‘’

The name if the Azure account name.

_This is mandatory if DEFAULT_FILESTORAGE=’storage.MultiAzureStorage’. Type: string

AZURE_CONTAINER:¶

Default: ‘’

The name if the Azure storage container.

_This is mandatory if DEFAULT_FILESTORAGE=’storage.MultiAzureStorage’. Type: string

AWS_ACCESS_KEY_ID¶

Default: ‘’

The host to use for polling emails.

Type: string

SCAN_EMAIL_HOST_USER¶

Default: ‘’

Username to use for the SMTP server defined in SCAN_EMAIL_HOST.

Type: string

SCAN_EMAIL_RECIPIENT¶

Default: ‘’

Recipient email address to be polled. Emails sent to this address will be polled into Konfuzio.

Type: string

SCAN_EMAIL_HOST_PASSWORD¶

Default: ‘’

The password to use for the SMTP server defined in SCAN_EMAIL_HOST.

Type: string

SCAN_EMAIL_SLEEP_TIME¶

Default: 60

The polling interval in seconds.

Default: 7200

The maximum time for the whole Document workflow in seconds. If a Document workflow does not complete within this time, the Document is set to an error state.

Default: RS256

Sets the algorithm the IdP uses to sign ID tokens. See https://mozilla-django-oidc.readthedocs.io/en/stable/settings.html#OIDC_RP_SIGN_ALGO

Type: string

9. Snapshot Settings (This feature is under active development and not yet available on https://app.konfuzio.com)¶

SNAPSHOT_DEFAULT_FILE_STORAGE¶

Default: ‘django.core.files.storage.FileSystemStorage’

By default, the file system is used as file storage. To use an S3-compatible storage set: ‘storages.backends.s3boto.S3BotoStorage’’. See https://docs.djangoproject.com/en/4.0/ref/settings/#default-file-storage

Type: string

SNAPSHOT_LOCATION¶

Default: ‘’

The location where all Snapshots are saved (e.g. “/data/konfuzio-snapshots”)

_This is mandatory if SNAPSHOT_DEFAULT_FILESTORAGE=’django.core.files.storage.FileSystemStorage’. Type: string _Type: string

These variables are only used for Keycloak integration tests: The admin variables are for login keycloak admin panel, the test variables are for login to Konfuzio server. The Keycloak Integration test can be run by installing Chromedriver and Selenium.

apt-get update && apt-get install -y fonts-liberation libatk-bridge2.0-0 libatk1.0-0 libatspi2.0-0 libgbm1 libgtk-3-0 libxcomposite1 libxkbcommon0 xdg-utils libu2f-udev
wget -q https://dl.google.com/linux/direct/google-chrome-stable_current_amd64.deb && apt-get install -y ./google-chrome-stable_current_amd64.deb
pip3 install coverage chromedriver-binary-auto chromedriver_binary
python manage.py test -t . server --tag=selenium-required

KEYCLOAK_ADMIN_USERNAME¶

Default: admin

The Keycloak admin username.

Type: string

KEYCLOAK_ADMIN_PASSWORD¶

Default: admin

The Keycloak admin password.

Type: string

KEYCLOAK_TEST_USERNAME¶

Default: fake@f.com

Username for Keycloak integration test.

Type: string

KEYCLOAK_TEST_PASSWORD¶

Default: pass1234

Password of User for Keycloak integration test.

Type: string

Example .env file for Konfuzio Server¶

Konfuzio Server is fully configured via environment variables, these can be passed as dedicated environment variables or a single .env to the Konfuzio Server containers (REGISTRY_URL/konfuzio/text-annotation/master). A template for a .env file is provided here:

# False for production, True for local development (mandatory).
# See https://docs.djangoproject.com/en/3.2/ref/settings/#std:setting-DEBUG
DEBUG=

# Set maintenance mode, shows 503 error page when maintenance-mode is on (mandatory).
MAINTENANCE_MODE=False

# Insert random secret key (mandatory).
# See https://docs.djangoproject.com/en/4.0/ref/settings/#secret-key
SECRET_KEY=

# The Billing API Key (optional)
BILLING_API_KEY=
# The URL of the biling Server (optional)
BILLING_API_URL=https://app.konfuzio.com

# The HOSTNAME variable is used in the E-Mail templates (mandatory).
# https://example.konfuzio.com or http://localhost:8000 for local development.
# Note: Please include the protocol (e.g. http://) even the variable is named HOST_NAME
HOST_NAME=

# Please enter a Postgres Database (https://github.com/kennethreitz/dj-database-url#url-schema) (mandatory).
DATABASE_URL=
DATABASE_URL_SSL_REQUIRE=True

# Insert hostname e.g. konfuzio.com or * for local development (mandatory).
# See https://docs.djangoproject.com/en/4.0/ref/settings/#allowed-hosts
ALLOWED_HOSTS=

# Django's default storage (mandatory).
# for azure use: storage.MultiAzureStorage
# for S3-like storage use: storage.MultiS3Boto3Storage
# See https://docs.djangoproject.com/en/4.0/ref/settings/#default-file-storage
DEFAULT_FILE_STORAGE=django.core.files.storage.FileSystemStorage

# Required settings if storage.MultiAzureStorage is used (optional).
AZURE_ACCOUNT_KEY=
AZURE_ACCOUNT_NAME=
AZURE_CONTAINER=

# Required settings for storage.MultiS3Boto3Storage (optional).
AWS_ACCESS_KEY_ID=
AWS_SECRET_ACCESS_KEY=
AWS_STORAGE_BUCKET_NAME=
AWS_S3_REGION_NAME=
AWS_S3_ENDPOINT_URL=

# Access to customer Blob Storage (optional).
# e.g. "{'beispiel_ag': {'account_key': 'the_keys_124','azure_container': 'default',}}"
AZURE_CUSTOMER_STORAGE={}

# Celery settings (mandatory).
BROKER_URL=
RESULT_BACKEND=
TASK_ALWAYS_EAGER=True

# Defender (Brute-Force protection) settings (optional).
DEFENDER_REDIS_URL=

# SENTRY_DSN e.g. "https://[email protected]/1234567" (optional).
SENTRY_DSN=

# E-Mail address which is BCC in every transactional E-Mail (optional).
CRM_INTEGRATION_EMAIL=

# The SMTP credentials for sending E-Mails (optional).
# See https://docs.djangoproject.com/en/4.0/ref/settings/#email-backend
EMAIL_BACKEND=django.core.mail.backends.smtp.EmailBackend
EMAIL_HOST=
EMAIL_HOST_PASSWORD=
EMAIL_HOST_USER=
EMAIL_PORT=25
# See https://docs.djangoproject.com/en/4.0/ref/settings/#email-use-tls
EMAIL_USE_TLS=False
# See https://docs.djangoproject.com/en/4.0/ref/settings/#email-use-ssl
EMAIL_USE_SSL=False
# See https://docs.djangoproject.com/en/4.0/ref/settings/#email-timeout
DEFAULT_FROM_EMAIL=

# Customize the email verification (optional)
# When set to “mandatory” the user is blocked from logging in until the email address is verified. Choose “optional” or “none” to allow logins with an unverified e-mail address. In case of “optional”, the e-mail verification mail is still sent, whereas in case of “none” no e-mail verification mails are sent.
ACCOUNT_EMAIL_VERIFICATION='mandatory'

# Api Key to sent emails via SendGrid if Debug=False (optional).
# If you use the SENDGRID_API_KEY you must also set EMAIL_BACKEND=sendgrid_backend.SendgridBackend
SENDGRID_API_KEY=

# Set Google Analytics or keep empty (optional).
GOOGLE_ANALYTICS=

# Captcha protected signup (optional).
CAPTCHA=False

# Flower URL or keep empty (optional).
FLOWER_URL=

# Rate limit per worker (optional).
TASK_DEFAULT_RATE_LIMIT=30/m

# Configure Azure OCR (optional).
AZURE_OCR_KEY=
AZURE_OCR_VERSION=
AZURE_OCR_BASE_URL=

# If this is activated SSL is required (optional).
# See https://docs.djangoproject.com/en/4.0/ref/settings/#std:setting-SESSION_COOKIE_SECURE
SESSION_COOKIE_SECURE=True

# If this is activated SSL is required (optional).
# https://docs.djangoproject.com/en/4.0/ref/settings/#csrf-cookie-secure
CSRF_COOKIE_SECURE=True

# New relic settings (optional).
NEW_RELIC_LICENSE_KEY=
NEW_RELIC_APP_NAME=
NEW_RELIC_ENVIRONMENT=

# Email integration (optional).
SCAN_EMAIL_HOST=
SCAN_EMAIL_HOST_USER=
SCAN_EMAIL_RECIPIENT=
SCAN_EMAIL_HOST_PASSWORD=

# Directory to cache files during the AI training process and when running AI models (optional).
KONFUZIO_CACHE_DIR =   # e.g. '/cache', uses tempdir if not set

# KEYCLOAK ENVIRONMENT The following values establish a keycloak connection through the
# mozilla oidc package (https://mozilla-django-oidc.readthedocs.io/en/stable/settings.html) (optional).

# SET TO TRUE TO ACTIVATE KEYCLOAK (optional).
SSO_ENABLED=True

# If you use keycloak version 17 and later set url like: http(s)://{keycloak_address}:{port}/ (optional).
# If you use keycloak version 16 and earlier set url like: http(s)://{keycloak_address}:{port}/auth/ (optional).
KEYCLOAK_URL=
KEYCLOAK_REALM=  # defaults to master

For Keycloak client creation see: https://www.keycloak.org/docs/latest/server_admin/#assembly-managing-clients_server_administration_guide (optional).
OIDC_RP_SIGN_ALGO=RS256
OIDC_RP_CLIENT_ID=
OIDC_RP_CLIENT_SECRET=

# These variables are only used for Keycloak integration tests:
# The admin variables are for login keycloak admin panel, the test variables are for login to Konfuzio server (optional).
KEYCLOAK_ADMIN_USERNAME=
KEYCLOAK_ADMIN_PASSWORD=
KEYCLOAK_TEST_USERNAME=
KEYCLOAK_TEST_PASSWORD=

# Turn on/off autoretraining (optional).
TRAIN_EXTRACTION_AI_AUTOMATICALLY_IF_QUEUE_IS_EMPTY=False

# Turn on/off the immediate generation of sandwich pdf in full document workflow (optional).
ALWAYS_GENERATE_SANDWICH_PDF=True

# Default time limits for background tasks (optional).
# These defaults can be viewed here: https://dev.konfuzio.com/web/explanations.html#celery-tasks.
EXTRACTION_TIME_LIMIT =
CATEGORIZATION_TIME_LIMIT =
EVALUATION_TIME_LIMIT =
TRAINING_EXTRACTION_TIME_LIMIT =
TRAINING_CATEGORIZATION_TIME_LIMIT =
SANDWICH_PDF_TIME_LIMIT =
DOCUMENT_TEXT_AND_BBOXES_TIME_LIMIT =
# Default time limits for period background tasks (optional)
# https://help.konfuzio.com/modules/projects/index.html?#auto-deletion-of-documents
# Both are set to 3600, the max amount of time the task may take.
# If a huge amount of documents have been deleted, this may need to be increased.
CLEAN_DELETED_DOCUMENT_TIME_LIMIT =
CLEAN_DOCUMENT_WITHOUT_DATASET_TIME_LIMIT =

# Some models require a lot of ram during training. Defaults to 150 threshold which is the amount of training documents necessary to switch to a "training_heavy" celery queue
THRESHOLD_FOR_HEAVY_TRAINING =

Environment Variables for Read API Container¶

 # The Azure OCR API key (mandatory).
AZURE_OCR_KEY=123456789
 # The URL of the READ API (mandatory).
AZURE_OCR_BASE_URL=http://host:5000
# The version of the READ API (optional).
AZURE_OCR_VERSION=v3.2

Environment Variables for Document Layout Analysis Container¶

# Connect Broker (mandatory).
BROKER_URL=
# Connect result backend (mandatory).
RESULT_BACKEND=
# Connect Sentry (optional).
SENTRY_DSN=
SENTRY_ENVIRONMENT=
SENTRY_RELEASE=
SENTRY_TRACES_SAMPLE_RATE=
# Setting for task processing (optional).
WORKER_SEND_TASK_EVENTS=
TASK_SEND_SENT_EVENT=
TASK_TRACK_STARTED=
TASK_ACKS_ON_FAILURE_OR_TIMEOUT=
TASK_ACKS_LATE=

Profiling¶

Django-silk is an integration into Konfuzio Server which offers a detailed overview of time spend in the database or within the internal code. When trying to troubleshoot performance issues, wanting to understand where queries are made, or needing to cooperate with our support on technical problems, it may be useful to use the Django-silk profiling to analyze performance.

By default, the package is installed but turned off, you can enable the package by setting the following variable:

SILK_PROFILING: True

By default, only 15% of requests are being profiled, this can be changed to any number between 0 and 100:

SILKY_INTERCEPT_PERCENT: 15

To avoid the database from filling up, only the past 10k requests are being kept, this can be changed to any value:

SILKY_MAX_RECORDED_REQUESTS: 10000

After enabling Django silk, the dashboard will become available on http://localhost:8000/silk for users who are logged in as superuser. The database can also be cleared by navigating to http://localhost:8000/silk/cleardb/. Do keep in mind that this package does add some overhead querying to each request. So unless there are active problems, it may be best to keep it turned off or at a low SILKY_INTERCEPT_PERCENT percentage.

Installation Guide¶

Billing and License¶

Setup Billing API¶

Important Updates¶

Technical Background¶

Air-gapped Environments and Upcoming Changes¶

Kubernetes¶

Required tools¶

kubectl¶

Helm¶

Getting Helm¶

Connect to a local Minikube cluster¶

Initializing Helm¶

Next steps¶

Deployment¶

Selecting configuration options¶

Secrets¶

Networking and DNS¶

Persistence¶

TLS certificates¶

PostgreSQL¶

Redis¶

Persistent Volume¶

CPU, GPU and RAM Resource Requirements¶

Storage Requirements¶

Deploy using Helm¶

Monitoring the Deployment¶

Autoscaling¶

Initial login¶

Quick Start via Kubernetes and Helm¶

Upgrade¶

Docker - Single VM setup¶

VM Requirements¶

Login to the Konfuzio Docker Image Registry¶

Quick Start via Docker-Compose¶

1. Download Docker Image¶

2. Setup PostgreSQL, Redis, BlobStorage/FileSystemStorage¶

3. Setup the environment variable file¶

4. Init the database, create first superuser via cli and prefill e-mail templates¶

5. Start the container¶

[Optional] 6. Use Flower to monitor tasks¶

[Optional] 7. Run Container for Email Integration¶

[Optional] 8. Use Azure Read API (On-Premises or as Service)¶

How to fix common Azure Read API issues for On-Premises¶

1. Ensure AVX2 Compatibility:¶

2. Verify Azure Read Container Status:¶

3. Consider Trying a Different Container Tag:¶

[Optional] 9. Install the Document Layout Analysis (Segmentation) Container¶

[Optional] 10. Install document summarization container¶

11a. Upgrade to newer Konfuzio Version¶

11b. Downgrade to older Konfuzio Version¶

Load Scenario for Single VM with 32GB¶

Scenario 1: With self-hosted OCR¶

Scenario 2: Without self-hosted OCR¶

Document Layout Analysis¶

Docker-Compose vs. Kubernetes¶

Custom AI model training via CI pipelines¶

SSO via OpenID Connect (OIDC)¶

SSO Keycloak Integration¶

Set up¶

Environment Variables¶

Important notes¶

SSO via other Identity Management software¶

Step 1: Setting up your Identity Provider¶

Step 2: Configuring mozilla-django-oidc¶

Step 3: Restarting Konfuzio¶

Step 4: Testing¶

Migrate AIs and Projects¶

Overview of Migration Methods¶

Migrate an Extraction or Categorization AI¶

Migrate a Project¶

Database and Storage¶

Overview¶

Usage of PostgreSQL¶

Environment Variables¶

Environment Variables for Konfuzio Server¶

1. General Settings¶

HOST_NAME¶

ALLOWED_HOSTS¶

BILLING_API_KEY¶