AI should be simple, safe, and amazing.
Data Concierge AI (aka Concierge) is an AI system that works ONLY with the data you feed it. Your data, your prompts, everything are local to your instance.
Concierge is a local and modular RAG framework still in alpha.
Built with simplicy and security in mind, it has some features we love -- and hope you do too!
These versions are currently being used to develop Concierge, lower versions may work but are untested. See Known Issues for an older Docker Compose command which might work for you if you're unable to upgrade.
python. On your system you may need to use python3, you can use the python-is-python3 package to configure the python command on Linux). Check with python --version.docker --version.docker compose version. Optional:
If you want to use GPU acceleration (Concierge does NOT require this, but it will make responses dramatically faster), you must have the
NVIDIA drivers correctly setup and running. Concierge will not install or make any adjustments to your driver configuration.
Note: if you want to use GPU acceleration on a Windows host, you must use WSL2.
More details here:
https://docs.docker.com/desktop/gpu/
Refer to the documentation from NVIDIA for information on how to do this for your OS.
It is unlikely you'll be able to run Concierge with less than 8GB RAM, at least 16GB is desirable.
If you're using Docker Desktop you need to make sure at least 4GB RAM is assigned to be used by containers. The default is half of the system memory, so if you have at least 8GB RAM and haven't modified your configuration it's likely you don't need to do anything.
You should perform the following system configuration steps according to your Operating System: https://opensearch.org/docs/latest/install-and-configure/install-opensearch/docker/#install-docker-and-docker-compose, otherwise you may not be able to run the OpenSearch container Concierge depends on.
The provided install scripts perform a lot of automated cleanup and ensure that all the components are configured correctly to work together, so we strongly recommend you use these instead of attempting the manual install. However if you need to customize the Docker environment the manual steps may help you find what you need to be able to do so.
You no longer need to clone this repository.
You have a couple of options depending on your preferences.
[!TIP] Pay attention to the use of dashes and underscores in the commands!
If you're not concerned about using a virtual environment you can just use the 2 commands below in the directory of your choosing:
python -m pip install launch-concierge
python -m launch_concierge.install
If you prefer to keep things contained in a virtual environment you can use the 4 commands below:
python -m venv .
.\scripts\activate.ps1 or source bin/activate depending on your Operating System
pip install launch-concierge
install_concierge
Answer the questions and then the installer will ask if you are ready to make changes to the system.
Answer "yes" and let the downloading begin!
Please note that the package isn't the Concierge app itself, it's just a utility that helps you configure the environment and launch the correct Docker Compose file based on your choices, so it shouldn't be hugely risky to download it without using a virtual environment.
If you would prefer to configure and launch the app without going through our utility script it is also fairly straightforward:
Create a file called .env following the template of .env.example in this repository.
From launcher_package/src/launch_concierge/docker_compose copy docker_compose_dependencies.
From the same directory, if you wish to use the CPU only, copy docker-compose.yml, if you wish to use GPU acceleration where available, copy docker-compose-gpu.yml instead.
To launch the CPU compose file use docker compose up -d, to launch the GPU version you need to use docker compose -f docker-compose-gpu.yml up -d.
Once you have set up the Docker containers using one of the methods above, Concierge will be running on localhost:15130 or the port you selected during setup. It can take a couple of minutes for the containers to be ready after install or relaunch.
If running a version prior to 0.3.0 you should delete the files you cloned from the repository, remove the related Docker containers and proceed with a fresh install following the instructions above.
python -m pip install launch-concierge --upgrade
python -m launch_concierge.install
Activate the environment if not already done
pip install launch-concierge --upgrade
install_concierge
Make sure to grab the latest version of the Docker Compose files.
docker compose pull or docker compose -f docker-compose-gpu.yml pull will get the latest versions of the containers being used.
Then you can simply launch the containers again using the command from the install step.
git clone repo or extract zip.
cd concierge go into the cloned project directory.
You should not create a virtual environment as the script below will handle it for you.
python install_dev.py to launch the installer (same steps as the user installer).
Configure the .env file like in the user install.
Set ENVIRONMENT to development
python -m venv . create a python virtual environment in the current directory.
Linux: source ./bin/activate / Windows PowerShell: .\Scripts\Activate.ps1 enter into the virtual environment.
pip install -r requirements.txt -r dev_requirements.txt install all dependencies.
pre-commit install install the linting script that hooks into the commit command.
docker compose -f docker-compose-dev.yml up -d will load the docker dependencies for developers.
docker compose -f docker-compose-dev-gpu.yml up -d will load the docker dependencies for developers and use the GPU.
If you want to build the code to a Docker container to simulate the production environment you can use the docker-compose-local.yml or docker-compose-gpu-local.yml files. On initial launch and when you've made changes to the code you'll need to use docker compose -f <docker-compose-file.yml> build to update the container and then launch the compose file again.
Complete one of the installation methods above.
Make sure to read the Contribution Guide to find out more about coding style enforcement and commit etiquette!
Install Shiny for Python VSCode extension.
Run Shiny for Python VSCode extension from concierge_shiny/app.py.
At the time of writing we have noticed an issue where the VSCode browser window doesn't automatically refresh and you have to copy/paste the URL from the console into it. Do this if after seeing the log Application startup complete. you still don't see anything in the VSCode browser.
You can also access the URL by pasting it into another browser.
From the cloned project directory simply run launch_dev.py.
If the Docker container dependencies aren't found, you will be given the option to launch with CPU or GPU.
python launch_local.py will build and launch the code in a Docker container as if it were the production environment.
If the above methods did not work:
If you are not in the python virtual environment, please enter it by the correct method:
Linux and MacOS: source ./bin/activate
Windows PowerShell: .\Scripts\Activate.ps1
To start the web UI, run the following command:
python -m shiny run --reload --launch-browser concierge_shiny/app.py
Clone the latest version of the repo.
Repeat the process used during install.
Use docker compose -f <docker-compose-file.yml> pull to update the Docker containers.
Repeat the process used during install.
While we're currently more focused on the GUI element, we have provided some CLI scripts to be able to perform some functions without launching the web app.
One notable feature is that the loader script loads an entire directory of documents which the GUI is currently unable to do.
To use them you can navigate to the cli subdirectory or append cli. to the script name from the parent directory.
Make sure you are running inside the venv.
Call commands like this: python -m <script_name> <options>. Use the -h or --help option to see what the options are.
Available commands:
loaderweb_loaderprompterlist_collectionslist_documentsdelete_collectiondelete_documentunknown shorthand flag: 'd' in -d and/or you have the docker-compose command instead of docker compose. This indicates that you're using an older version of Docker than we support. The best course of action would be to install the latest version following instructions from here: https://docs.docker.com/engine/install/. However if you're unable to do this, you may be able to get the Concierge Docker requirements running using docker-compose --file ./docker-compose.yml up.NotOpenSSLWarning, as far as we aware you can ignore this warning without issue. This should only be present when using the development environment.Please read our Code of Conduct then our Contribution Guide.