📦

LLM Server Docs

End-to-end documentation to set up your own local & fully private LLM server on Debian. Equipped with chat, web search, RAG, model management, MCP servers, image generation, and TTS.

0 installs

Trust: 59 — Fair

Rag

Ask AI about LLM Server Docs

I know everything about LLM Server Docs. Ask me about installation, configuration, usage, or troubleshooting.

0/500

Loading tools...

Reviews

Documentation

Local LLaMA Server Setup Documentation

TL;DR: End-to-end documentation to set up your own local & fully private LLM server on Debian. Equipped with chat, web search, RAG, model management, MCP servers, image generation, and TTS, along with steps for configuring SSH, firewall, and secure remote access via Tailscale.

Software Stack:

Inference Engine (Ollama, llama.cpp, vLLM)
Search Engine (SearXNG)
Model Server (llama-swap, systemd service)
Chat Platform (Open WebUI)
MCP Proxy Server (mcp-proxy, MCPJungle)
Text-to-Speech Server (Kokoro FastAPI)
Image Generation Server (ComfyUI)

Software Stack Architectural Diagram

Local LLaMA Server Setup Documentation

About

This repository outlines the steps to run a server for running local language models. It uses Debian specifically, but most Linux distros should follow a very similar process. It aims to be a guide for Linux beginners like me who are setting up a server for the first time.

The process involves installing the requisite drivers, setting the GPU power limit, setting up auto-login, and scheduling the init.bash script to run at boot. All these settings are based on my ideal setup for a language model server that runs most of the day but a lot can be customized to suit your needs.

[!IMPORTANT] No part of this guide was written using AI - any hallucinations are the good old human kind. While I've done my absolute best to ensure correctness in every step/command, check everything you execute in a terminal. Enjoy!

Priorities

Simplicity: It should be relatively straightforward to set up the components of the solution.
Stability: The components should be stable and capable of running for weeks at a time without any intervention necessary.
Security: The components should be able to be tightly secured and limited in their capability to damage the system in case of a known vulnerability affecting any of the components.
Maintainability: The components and their interactions should be uncomplicated enough that you know enough to maintain them as they evolve (because they will evolve).
Aesthetics: The result should be as close to a cloud provider's chat platform as possible. A homelab solution doesn't necessarily need to feel like it was cobbled together haphazardly.
Modularity: Components in the setup should be able to be swapped out for newer/more performant/better maintained alternatives easily. Standard protocols (OpenAI-compatibility, MCPs, etc.) help with this a lot and, in this guide, they are always preferred over bundled solutions.
Open source: The code should be able to be verified by a community of engineers. Chat platforms and LLMs involve large amounts of personal data conveyed in natural language and it's important to know that data isn't going outside your machine.

Prerequisites

Any modern CPU and GPU combination should work for this guide. Previously, compatibility with AMD GPUs was an issue but the latest releases of Ollama have worked through this and AMD GPUs are now supported natively.

For reference, this guide was built around the following system:

CPU: Intel Core i5-12600KF
Memory: 96GB 3200MHz DDR4 RAM
Storage: 1TB M.2 NVMe SSD
GPU: 2x Nvidia RTX 3090 (24GB)

[!NOTE] AMD GPUs: Power limiting is skipped for AMD GPUs as AMD has recently made it difficult to set power limits on their GPUs. Naturally, skip any steps involving nvidia-smi or nvidia-persistenced and the power limit in the init.bash script.

CPU-only: You can skip the GPU driver installation and power limiting steps. The rest of the guide should work as expected.

[!NOTE] This guide uses ~/ (or /home/<your_username>) as the base directory. If you're working in different directory, please modify all your commands accordingly.

To begin the process of setting up your server, you will need the following:

Fresh install of Debian
Internet connection
Basic understanding of the Linux terminal
Peripherals like a monitor, keyboard, and mouse

To install Debian on your newly built server hardware:

Download the Debian ISO from the official website.
Create a bootable USB using a tool like Rufus for Windows or Balena Etcher for MacOS.
Boot into the USB and install Debian.

For a more detailed guide on installing Debian, refer to the official documentation. For those who aren't yet experienced with Linux, I recommend using the graphical installer - you will be given an option between the text-based installer and graphical installer.

I also recommend installing a lightweight desktop environment like XFCE for ease of use. Other options like GNOME or KDE are also available - GNOME may be a better option for those using their server as a primary workstation as it is more feature-rich (and, as such, heavier) than XFCE.

General

Allow `sudo` Permissions

To do a bunch of things in this guide, we need to be root. But we don't have the ability to act as root until root allows our user to. First, we'll switch to root and grant our user permission to run commands with sudo.

[!TIP] sudo stands for "superuser doer" - in Linux, it signals to the OS that you want the command you're running to be running on behalf of the root user. This should be used sparingly on a highly secure system (we want user-specific permissions for most processes) and carefully since it can affect the system with the same power as the root user. Don't overthink its use or worry when using it - just know that it can be dangerous if used incorrectly.

Switch to root:
```
su root
```
Run the following command to add your user to the sudo group (which has the permissions we're looking for):
```
sudo usermod -a -G sudo <username>
```
Replace <username> with your username.

Save and exit (Ctrl+X).
Close your existing terminal and open a new session. This is required to see the changes.
(Optional) Test your new permissions by running ls (command to view the contents of a directory) with sudo:
```
sudo ls
```

Update System Packages

Update the system by running the following commands:
```
sudo apt update
sudo apt upgrade
```

Now, we'll install the required GPU drivers that allow programs to utilize their compute capabilities.

Nvidia GPUs

Follow Nvidia's guide on downloading CUDA Toolkit. The instructions are specific to your machine and the website will lead you to them interactively.

Run the following commands:

sudo apt install linux-headers-amd64
sudo apt install nvidia-driver firmware-misc-nonfree

Reboot the server.
Run the following command to verify the installation:
```
nvidia-smi
```

AMD GPUs

Run the following commands:

deb http://deb.debian.org/debian bookworm main contrib non-free-firmware
apt install firmware-amd-graphics libgl1-mesa-dri libglx-mesa0 mesa-vulkan-drivers xserver-xorg-video-all

Reboot the server.

We'll also install some packages that are not installed on Debian by default but may be required later:

sudo apt install libcurl cmake

Schedule Startup Script

In this step, we'll create a script called init.bash. This script will be run at boot to set the GPU power limit and start the server using Ollama. We set the GPU power limit lower because it has been seen in testing and inference that there is only a 5-15% performance decrease for a 30% reduction in power consumption. This is especially important for servers that are running 24/7.

Run the following commands:
```
touch init.bash
nano init.bash
```
Add the following lines to the script:
```
#!/bin/bash
sudo nvidia-smi -pm 1
sudo nvidia-smi -pl <power_limit>
```
Replace <power_limit> with the desired power limit in watts. For example, sudo nvidia-smi -pl 250.

For multiple GPUs, modify the script to set the power limit for each GPU:
```
sudo nvidia-smi -i 0 -pl <power_limit>
sudo nvidia-smi -i 1 -pl <power_limit>
```
Save and exit the script.
Make the script executable:
```
chmod +x init.bash
```

Adding the init.bash script to the crontab will schedule it to run at boot.

Run the following command:
```
crontab -e
```
Add the following line to the file:
```
@reboot /path/to/init.bash
```
Replace /path/to/init.bash with the path to the init.bash script.
(Optional) Add the following line to shutdown the server at 12am:
```
0 0 * * * /sbin/shutdown -h now
```
Save and exit the file.

Configure Script Permissions

We want init.bash to run the nvidia-smi commands without having to enter a password. This is done by giving nvidia-persistenced and nvidia-smi passwordless sudo permissions, and can be achieved by editing the sudoers file.

AMD users can skip this step as power limiting is not supported on AMD GPUs.

Run the following command to edit the sudoers file:
```
sudo visudo
```

Add the following lines to the file:

<username> ALL=(ALL) NOPASSWD: /usr/bin/nvidia-persistenced
<username> ALL=(ALL) NOPASSWD: /usr/bin/nvidia-smi

Replace <username> with your username.

Save and exit the file.

[!IMPORTANT] Ensure that you add these lines AFTER %sudo ALL=(ALL:ALL) ALL. The order of the lines in the file matters - the last matching line will be used so if you add these lines before %sudo ALL=(ALL:ALL) ALL, they will be ignored.

Configure Auto-Login (optional)

When the server boots up, we may want it to automatically log in to a user account and run the init.bash script. This is done by configuring the lightdm display manager.

Run the following command:
```
sudo nano /etc/lightdm/lightdm.conf
```
Find the following commented line. It should be in the [Seat:*] section.
```
# autologin-user=
```
Uncomment the line and add your username:
```
autologin-user=<username>
```
Replace <username> with your username.
Save and exit the file.

Docker

📖 Documentation

Docker is a containerization platform that allows you to run applications in isolated environments. This subsection follows Docker's guide to install Docker Engine on Debian. The commands are listed below, but visiting the guide is recommended in case instructions have changed.

If you already have a Docker installation on your system, it's a good idea to re-install so there are no broken/out-of-date dependencies. The command below will iterate through your system's installed packages and remove the ones associated with Docker.
```
for pkg in docker.io docker-doc docker-compose podman-docker containerd runc; do sudo apt purge $pkg; done
```

Run the following commands:

# Add Docker's official GPG key:
sudo apt update
sudo apt install ca-certificates curl
sudo install -m 0755 -d /etc/apt/keyrings
sudo curl -fsSL https://download.docker.com/linux/debian/gpg -o /etc/apt/keyrings/docker.asc
sudo chmod a+r /etc/apt/keyrings/docker.asc

# Add the repository to Apt sources:
sudo tee /etc/apt/sources.list.d/docker.sources <<EOF
Types: deb
URIs: https://download.docker.com/linux/debian
Suites: $(. /etc/os-release && echo "$VERSION_CODENAME")
Components: stable
Signed-By: /etc/apt/keyrings/docker.asc
EOF

sudo apt update

Install the Docker packages:

sudo apt install docker-ce docker-ce-cli containerd.io docker-buildx-plugin docker-compose-plugin

Verify the installation (optional):
```
sudo systemctl status docker
```
If it hasn't started, try manually starting the daemon:
```
sudo systemctl start docker
```

Add User to Docker Group

So that we can forgo the use of sudo for Docker commands and run them directly under our user, we will add our user to the docker group. This isn't strictly necessary but convenient. This sub-section uses Docker's own post-install documentation as a reference - have a look just in case things have changed since the time of writing.

[!CAUTION] Being in the docker group grants significant system access - effectively root-equivalent privileges to the Docker daemon. Only add trusted users to this group.

Add your user to the Docker group:
```
sudo usermod -aG docker $USER
```
Apply the changes:
```
newgrp docker
```
Verify your new permissions by checking running containers:
```
docker ps -a
```

If you encounter "permission denied" and Docker commands still require sudo, log out and log back in.

[!NOTE] If you decide not to do this, you will need to append sudo in front of any docker commands.

Nvidia Container Toolkit

You will most likely want to use GPUs via Docker - this will require Nvidia Container Toolkit, which allows Docker to allocate/de-allocate memory on Nvidia GPUs. The steps for installing this are listed below, but it is recommended to reference Nvidia's documentation for the most up-to-date commands.

Configure the repository

curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey | sudo gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg \
&& curl -s -L https://nvidia.github.io/libnvidia-container/stable/deb/nvidia-container-toolkit.list | \
    sed 's#deb https://#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://#g' | \
    sudo tee /etc/apt/sources.list.d/nvidia-container-toolkit.list

Update packages:
```
sudo apt update
```

Install Nvidia Container Toolkit packages:

sudo apt install -y nvidia-container-toolkit

Restart the Docker daemon:
```
sudo systemctl restart docker
```

Create a Network

We'll be running most services via Docker containers. To allow multiple containers to communicate with each other, we can open up ports via UFW (which we'll configure later) but this is less optimal than creating a Docker network. This way, all containers on the network can securely talk to each other without needing to open ports for the many services we could run via UFW, inherently creating a more secure setup.

We'll call this network app-net: you can call it anything you like, just be sure to update the commands that use it later.

Run the following:

docker network create app-net

That's it! Now, when we create containers, we can reference it as follows:

Docker Run

docker run <container> --network app-net

Docker Compose

services:
  <container>:
    # add this
    networks:
      - app-net

# add this
networks:
  app-net:
    external: true

Replace <container> with the actual service - don't forget to add other parameters too.

With this configured, we can now call containers by name and port. Let's pretend we're calling the /health endpoint in llama-swap from open-webui (two actual containers we'll be creating later on) to ensure that the containers can see and speak to each other. Run (CTRL+C to quit):

docker exec -i open-webui curl http://llama-swap:8080/health

You could also run it the other way to be extra sure:

docker exec -it llama-swap curl http://open-webui:8080

[!IMPORTANT] The port is always the internal port the container is running on. If a container runs on 1111:8080, for example, 1111 is the port on the host (where you might access it, like http://localhost:1111 or http://<server_ip>:1111) and 8080 is the internal port the container is running on. Thus, trying to access the container on 1111 via app-net will not work. Remembering this when specifying URLs in services will save you a lot of unnecessary "why isn't this working?" pains.

Harden Docker Containers

To "harden" something in software terms is to make it more secure and more resilient to cyberattacks. Usually, this involves reducing the surface area by which an attacker can gain access to the system but it also includes mitigating what an attacker can do assuming they have successfully gotten access. Essentially, we will set up both prevention and cure - although, as they say, an ounce of prevention is better than a pound of cure. This sub-section leverages knowledge from this Reddit comment, courtesy of u/arnedam. Another good reference is the OWASP Docker Security Cheat Sheet.

To a large extent, the security objectives we're aiming towards can be achieved by applying the Principle of Least Privilege (PoLP) - in other words, giving our containers exactly and only what they need to perform their function. For each guideline, I've provided an existing risk and a mitigation strategy (along with how it fixes the problem); while not strictly necessary, learning how things are at risk is beneficial to start thinking securely about the services you host.

[!NOTE] This sub-section is critical for public-facing services. However, even if you only intend to access your services via a private network (preferred if possible), doing this is quite literally never a bad idea because it limits how much damage can be done to your server and its services if it ever was to become compromised somehow.

Run the service under your user, not root:
```
user: "<your_user_id>:<your_group_id>"
```
To find <your_user_id>, run id -u. To find <your_group_id>, run id -g.

RISK: Unless specified otherwise, containers run under the root user. Compromised containers running under root can do all sorts of damage, ranging anywhere from executing malicious code to prying on/deleting system files.

MITIGATION: By specifying a user that the container runs under, the container's privilege is limited to the user's privilege, which is almost definitively lower than the root user's privilege to modify the system.
Make the file system read-only:
```
read_only: true
```
RISK: By default, containers are allowed to read and write the files they have access to. This means an attacker who has successfully compromised your container could write bogus files/malicious code and spy on/delete critical information.

MITIGATION: Adding this will prevent attackers from being able to write malicious code files or delete important data from your system that are accessible to the container.

[!WARNING] Apply the read_only: true directive with care: if your service leverages file writes, like the many services that write to /tmp, those subprocesses will fail. If those subprocesses are critical, you may even see the container fail to start. Expect this not to work in an all-encompassing way for all containers - you will most likely need to surgically apply the :ro directive to volume mounts in most places.

Limit the physical resources your container can access:
```
# maximum number of processes the container can execute
pids_limit: 512
# maximium memory for the container
mem_limit: 3g
# maximum number of CPU cores the container can utilize - can be fractional
cpus: 3
```
RISK: Containers are, by default, given unrestricted physical resource allocation capabilities. This means that if a container was to be compromised, it would be able to allocate as much of your CPU clock cycles, RAM capacity, and VRAM capacity to executing malicious code.

MITIGATION: Allocating fixed limits to resources will prevent attackers from using the entirety of your server's physical resources to do damage if your container was to be compromised, e.g. running a botnet via your machine.

Docker Swarm

If you're using Docker Swarm (not used in this guide), you need to format it slightly differently:
```
deploy:
  resources:
    limits:
      pids: 512
      memory: 3g
      cpus: 3
```
Disable tty and stdin in the container:
```
tty: false
stdin_open: false
```
RISK: stdin gives the container the ability for commands to be written to it (input injection) and tty gives the container an active shell environment - together, they grant the container complete interactive shell capability such that potentially malicious commands can be run from it.

MITIGATION: Disabling these will prevent attackers from being able to execute code via the containers, severely minimizing arbitrary code execution (ACE) vulnerabilities.
Disallow the container from elevating its own privileges:
```
security_opt:
  - "no-new-privileges=true"
```
RISK: Containers are able to elevate their own access given an interactive shell environment. This could even override the user directive provided earlier, giving root-level access to the system via the container.

MITIGATION: Adding this line will prevent attackers from being able to override the user directive we provided earlier and stop them from being able to grant the now-compromised container root-level permissions.
Drop default container capabilities:
```
cap_drop:
  - ALL
```
RISK: Containers are given a lot of permissions by default. Most of these permissions are not required by most containers. Giving extra access increases the attack vector surface area in case a container was to become compromised - especially since these capabilities can impact the host kernel.

MITIGATION: Adding this will drop the widely-permissive default permissions granted to containers. This ensures that we only give containers the system capabilities to do the things they are required to do.

cap_drop with a value of ALL can be too aggressive for containers that require multiple capabilities. In this case, right after the above block, consider reviewing the following commonly-used capabilities. They are not all required and should not all be added - they are just a starting point for some capabilities sometimes required by containers.
```
cap_add:
  # LOWER RISK
  # Make arbitrary changes to file UIDs and GIDs (see chown(2))
  - CHOWN
  # Make arbitrary manipulations of process GIDs and supplementary GID list
  - SETGID
  # Make arbitrary manipulations of process UIDs
  - SETUID
  # Bind a socket to internet domain privileged ports (port numbers less than 1024)
  - NET_BIND_SERVICE

  # HIGHER RISK - CAN AFFECT IMPORTANT SYSTEM-LEVEL CONFIGURATIONS
  # Perform various network-related operations
  - NET_ADMIN
  # Use RAW and PACKET sockets
  - NET_RAW
  # Perform a range of system administration operations
  - SYS_ADMIN
```
This will, in order, drop all capabilities and then surgically re-add the ones that are required for the functionality the container achieves. A complete list of capabilities can be found here.

Containers with inadequate capabilities will fail to run. If you drop one that is required, you could:
1. Ideally, take a look at the list of capabilities referenced above, and perform some trial-and-error addition of capabilities. LLMs can help greatly with this endeavor, especially when equipped with tools that can make them capable of reading documentation or the source code of the service you're looking to run. You can skip this directive until you have everything configured and come back to it later.
2. Less ideally, give up and remove the cap_drop directive entirely. I wouldn't recommend it but, with private services, this is hardly the most insecure setup in the world if you do everything else.
Prevent excessive logging:
```
logging:
  driver: "json-file"
  options:
    max-file: "10"
    max-size: "20m"
```
RISK: Much like a zip bomb, a logging bomb can recursively clutter and eventually render the system unusable by overwhelming the system with an ever-increasing number of large log files.

MITIGATION: By capping the maximum number of files (10) and sizes of the individual files (20MB), we can limit the effectiveness of an attack like this.

Feel free to change the limits (reasonably) as you wish if you run into issues.
Limit the temporary file directory's privileges:
```
tmpfs:
  - /tmp:rw,noexec,nosuid,nodev,size=512m
```
RISK: If allowed to download files, compromised containers could both install malicious files to the filesystem and execute them.

MITIGATION: Setting up a temporary file area with very limited permissions stops this potential attack vector. Containers are limited to downloading files here, in a little sandbox with a fixed small size and without script execution privileges.
Mount read-only volumes:
```
volumes:
  - /path/to/mount1:/mount1:ro
  - /path/to/mount2:/mount2:ro
```
RISK: To access the file system of the host, Docker containers need directories "mounted" as volumes. This lets them read and write from the host as if the container was the host itself. Allowing write access to every mount is usually not necessary and can allow a compromised container to delete or overwrite important information in a cyber-attack.

MITIGATION: Mounting volumes as read-only, wherever possible, eliminates the container's ability to destroy the data in those volumes via writes. This wouldn't prevent spyware, but it would limit data from being lost in an attack.

Replace /path/to/mount1 and /path/to/mount2 with actual directory paths.

[!IMPORTANT] For containers where free allocation of CPU cores and memory is crucial - llama-swap, for example - you will not want to limit the maximum resources the container can access as shown in step 3.

Here's a combined chunk to copy + paste into your existing services' Compose files:

services:
  <service_name>:
    # 1
    user: "<your_user_id>:<your_group_id>"
    # 2 - Uncomment only if the container does not write
    # read_only: true
    # 3
    pids_limit: 512
    mem_limit: 3g
    cpus: 3
    # 4
    tty: false
    stdin_open: false
    # 5
    security_opt:
      - "no-new-privileges=true"
    # 6
    cap_drop:
      - ALL
    # Add cap_add section if required
    # 7
    logging:
      driver: "json-file"
      options:
        max-file: "10"
        max-size: "20m"
    # 8
    tmpfs:
      - /tmp:rw,noexec,nosuid,nodev,size=512m
    # 9
    volumes:
      - /path/to/mount1:/mount1:ro
      - /path/to/mount2:/mount2:ro

Helpful Commands

In the process of setting up this server (or anywhere down the rabbit hole of setting up services), you're likely to use Docker often. For the uninitiated, here are helpful commands that will make navigating and troubleshooting containers easier:

See available/running containers: docker ps -a
Restart a container: docker restart <container_name>
View a container's logs (live): docker logs -f <container_name> (CTRL+C to quit)
Rename a container: docker rename <container_name> <new_container_name>
Sometimes, a single service will spin up multiple containers, e.g. xyz-server and xyz-db. To restart both simultaneously, run the following command from inside the directory containing the Compose file: docker compose restart
Recreate a service: docker compose down && docker compose up -d
Test a container's configuration: docker compose config --quiet
View a container's resolved configuration: docker compose config
List Docker networks: docker network ls

[!TIP] There are no set rules when it comes to how you set up your Docker containers/services. However, here are my two cents:
It's cleanest to use Docker Compose (docker compose up -d with a docker-compose.yaml file as opposed to docker run -d <image_name>). Unless you take copious notes on your homelab and its setup, this method is almost self-documenting and keeps a neat trail of the services you run via their compose files. One compose file per directory is standard.

[!TIP] Occasionally, restarting a container is not enough to persist changes to configurations. If ever you find that changes you made to a service are not showing up, recreate the service altogether first before continuing to other troubleshooting.

HuggingFace CLI

📖 Documentation

HuggingFace is the leading open-source ML/AI platform - it hosts models (including LLMs), datasets, and demo apps that can be used to test models. For the purpose of this guide, we'll be using HuggingFace to download popular open-source LLMs.

[!NOTE] Only needed for llama.cpp/vLLM.

Create a new virtual environment:

python3 -m venv hf-env
source hf-env/bin/activate

Download the huggingface_hub package using pip:
```
pip install -U "huggingface_hub[cli]"
```
Create an authentication token on https://huggingface.com
Log in to HF Hub:
```
hf auth login
```
Enter your token when prompted.
Run the following to verify your login:
```
hf auth whoami
```
The result should be your username.

Manage Models

Models can be downloaded either to the default location (.cache/huggingface/hub) or to any local directory you specify. Where the model is stored can be defined using the --local-dir command line flag. Not specifying this will result in the model being stored in the default location. Storing the model in the folder where the packages for the inference engine are stored is good practice - this way, everything you need to run inference on a model is stored in the same place. However, if you use the same models with multiple backends frequently (e.g. using Qwen_QwQ-32B-Q4_K_M.gguf with both llama.cpp and vLLM), either set a common model directory or use the default HF option without specifying this flag.

First, activate the virtual environment that contains huggingface_hub:

source hf-env/bin/activate

Download Models

Models are downloaded using their HuggingFace tag. Here, we'll use bartowski/Qwen_QwQ-32B-GGUF as an example. To download a model, run:

hf download bartowski/Qwen_QwQ-32B-GGUF Qwen_QwQ-32B-Q4_K_M.gguf --local-dir models

Ensure that you are in the correct directory when you run this.

Delete Models

To delete a model in the specified location, run:

rm <model_name>

To delete a model in the default location, run:

hf delete-cache

This will start an interactive session where you can remove models from the HuggingFace directory. In case you've been saving models in a different location than .cache/huggingface, deleting models from there will free up space but the metadata will remain in the HF cache until it is deleted properly. This can be done via the above command but you can also simply delete the model directory from .cache/huggingface/hub.

Search Engine

[!NOTE] This step is optional but highly recommended for grounding LLMs with relevant search results from reputable sources. Targeted web searches via MCP tool calls make reports generated by LLMs much less prone to random hallucinations.

SearXNG

🌟 GitHub
📖 Documentation

To power our search-based workflows, we don't want to rely on a search provider that can monitor searches. While using any search engine has this problem, metasearch engines like SearXNG mitigate it to a decent degree. SearXNG aggregates results from over 245 search services and does not track/profile users. You can use a hosted instance on the Internet but, considering the priorities of this guide and how trivial it is to set one up, we'll be spinning up our own instance on port 5050.

Start the container:

docker pull searxng/searxng
export PORT=5050
docker run \
    -d -p ${PORT}:8080 \
    --name searxng \
    --network app-net \
    -v "${PWD}/searxng:/etc/searxng" \
    -e "BASE_URL=http://0.0.0.0:$PORT/" \
    -e "INSTANCE_NAME=searxng" \
    --restart unless-stopped \
    searxng/searxng

Edit settings.yml to include JSON format support:

sudo nano searxng/settings.yml

Add the following:

  search:
    # ...other parameters here...
    formats:
        - html
        - json      # add this line

Restart the container with docker restart searxng

Open WebUI Integration

In case you want a simple web search workflow and want to skip MCP servers/agentic setups, Open WebUI supports web search functionality natively. Navigate to Admin Panel > Settings > Web Search and set the following values:

Enable Web Search
Web Search Engine: searxng
Searxng Query URL: http://searxng:8080/search?q=<query>
API Key: anything-you-like

Inference Engine

The inference engine is one of the primary components of this setup. It is code that takes model files containing weights and makes it possible to get useful outputs from them. This guide allows a choice between llama.cpp, vLLM, and Ollama - all of these are popular inference engines with different priorities and stengths (note: Ollama uses llama.cpp under the hood and is simply a CLI wrapper). It can be daunting to jump straight into the deep end with command line arguments in llama.cpp and vLLM. If you're a power user and enjoy the flexibility afforded by tight control over serving parameters, using either llama.cpp or vLLM will be a wonderful experience and really come down to the quantization format you decide. However, if you're a beginner or aren't yet comfortable with this, Ollama can be convenient stopgap while you build the skills you need or the very end of the line if you decide your current level of knowledge is enough!

Ollama

🌟 GitHub
📖 Documentation
🔧 Engine Arguments

Ollama will be installed as a service, so it runs automatically at boot.

Download Ollama from the official repository:

curl -fsSL https://ollama.com/install.sh | sh

We want our API endpoint to be reachable by the rest of the LAN. For Ollama, this means setting OLLAMA_HOST=0.0.0.0 in the ollama.service.

Run the following command to edit the service:
```
systemctl edit ollama.service
```
Find the [Service] section and add Environment="OLLAMA_HOST=0.0.0.0" under it. It should look like this:
```
[Service]
Environment="OLLAMA_HOST=0.0.0.0"
```
Save and exit.

Reload the environment.

systemctl daemon-reload
systemctl restart ollama

[!TIP] If you installed Ollama manually or don't use it as a service, remember to run ollama serve to properly start the server. Refer to Ollama's troubleshooting steps if you encounter an error.

llama.cpp

🌟 GitHub
📖 Documentation
🔧 Engine Arguments

Clone the llama.cpp GitHub repository:

git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp

Build the binary:

CPU
```
cmake -B build
cmake --build build --config Release
```
CUDA
```
cmake -B build -DGGML_CUDA=ON
cmake --build build --config Release
```
For other systems looking to use Metal, Vulkan and other low-level graphics APIs, view the complete llama.cpp build documentation to leverage accelerated inference.

vLLM

🌟 GitHub
📖 Documentation
🔧 Engine Arguments

vLLM comes with its own OpenAI-compatible API that we can use just like Ollama. Where Ollama runs GGUF model files, vLLM can run AWQ, GPTQ, GGUF, BitsAndBytes, and safetensors (the default release type) natively.

Manual Installation (Recommended)

Create a directory and virtual environment for vLLM:

mkdir vllm
cd vllm
python3 -m venv .venv
source .venv/bin/activate

Install vLLM using pip:
```
pip install vllm
```
Serve with your desired flags. It uses port 8000 by default, but I'm using port 8556 here so it doesn't conflict with any other services:
```
vllm serve <model> --port 8556
```
To use as a service, add the following block to init.bash to serve vLLM on startup:
```
source .venv/bin/activate
vllm serve <model> --port 8556
```
Replace <model> with your desired model tag, copied from HuggingFace.

Docker Installation

Run:

docker run --gpus all \
-v ~/.cache/huggingface:/root/.cache/huggingface \
--env "HUGGING_FACE_HUB_TOKEN=<your_hf_hub_token>" \
-p 8556:8000 \
--ipc=host \
vllm/vllm-openai:latest \
--model <model>

Replace <your_hf_hub_token> with your HuggingFace Hub token and <model> with your desired model tag, copied from HuggingFace.

To serve a different model:

First stop the existing container:

docker ps -a
docker stop <vllm_container_ID>

If you want to run the exact same setup again in the future, skip this step. Otherwise, run the following to delete the container and not clutter your Docker container environment:
```
docker rm <vllm_container_ID>
```

Rerun the Docker command from the installation with the desired model.

docker run --gpus all \
-v ~/.cache/huggingface:/root/.cache/huggingface \
--env "HUGGING_FACE_HUB_TOKEN=<your_hf_hub_token>" \
-p 8556:8000 \
--ipc=host \
vllm/vllm-openai:latest \
--model <model>

Open WebUI Integration

[!NOTE] Only needed for llama.cpp/vLLM.

Navigate to Admin Panel > Settings > Connections and set the following values:

Enable OpenAI API
API Base URL: http://host.docker.internal:<port>/v1
API Key: anything-you-like

[!NOTE] host.docker.internal is a magic hostname that resolves to the internal IP address assigned to the host by Docker. This allows containers to communicate with services running on the host, such as databases or web servers, without needing to know the host's IP address. It simplifies communication between containers and host-based services, making it easier to develop and deploy applications.

Ollama vs. llama.cpp

Aspect	Ollama (Wrapper)	llama.cpp (Vanilla)
Installation/Setup	One-click install & CLI model management	Requires manual setup/configuration
Open WebUI Integration	First-class citizen	Requires OpenAI-compatible endpoint setup
Model Switching	Native model-switching via server	Requires manual port management or llama-swap
Customizability	Limited: Modelfiles are cumbersome	Full control over parameters via CLI
Transparency	Defaults may override model parameters (e.g., context length)	Full transparency in parameter settings
GGUF Support	Inherits llama.cpp's best-in-class implementation	Best GGUF implementation
GPU-CPU Splitting	Inherits llama.cpp's efficient splitting	Trivial GPU-CPU splitting out-of-the-box

vLLM vs. Ollama/llama.cpp

Feature	vLLM	Ollama/llama.cpp
Vision Models	Supports Qwen 2.5 VL, Llama 3.2 Vision, etc.	Ollama supports some vision models, llama.cpp does not support any (via llama-server)
Quantization	Supports AWQ, GPTQ, BnB, etc.	Only supports GGUF
Multi-GPU Inference	Yes	Yes
Tensor Parallelism	Yes	No

In summary,

Ollama: Best for those who want an "it just works" experience.
llama.cpp: Best for those who want total control over their inference servers and are familiar with engine arguments.
vLLM: Best for those who want (i) to run non-GGUF quantizations of models, (ii) multi-GPU inference using tensor parallelism, or (iii) to use vision models.

Using Ollama as a service offers no degradation in experience because unused models are offloaded from VRAM after some time. Using vLLM or llama.cpp as a service keeps a model in memory, so I wouldn't use this alongside Ollama in an automated, always-on fashion unless it was your primary inference engine. Essentially,

Primary Engine	Secondary Engine	Run SE as service?
Ollama	llama.cpp/vLLM	No
llama.cpp/vLLM	Ollama	Yes

Model Server

[!NOTE] Only needed for manual installations of llama.cpp/vLLM. Ollama manages model management via its CLI.

While the above steps will help you get up and running with an OpenAI-compatible LLM server, they will not help with this server persisting after you close your terminal window or restart your physical server. They also won't allow a chat platform to reliably reference and swap between the various models you have available - a likely use-case in a landscape where different models specialize in different tasks. Running the inference engine via Docker can achieve this persistence with the -d (for "detach") flag but (i) services like llama.cpp and vLLM are usually configured without Docker and (ii) it can't swap models on-demand. This necessitates a server that can manage loading/unloading, swapping, and listing available models.

llama-swap

🌟 GitHub
📖 Documentation

[!TIP] This is my recommended way to run llama.cpp/vLLM models.

llama-swap is a lightweight proxy server for LLMs that solves our pain points from above. It's an extremely configurable tool that allows a single point of entry for models from various backends. Models can be set up in groups, listed/unlisted easily, configured with customized hyperparameters, and monitored using streamed logs in the llama-swap web UI.

In the installation below, we'll use Qwen3-4B-Instruct-2507-UD-Q4_K_XL.gguf for llama.cpp and Qwen/Qwen3-4B-Instruct-2507 for vLLM. We'll also use port 7000 to serve the models on.

Create a new directory with the config.yaml file:

sudo mkdir llama-swap
cd llama-swap
sudo nano config.yaml

Enter the following and save:

llama.cpp

models:
    "qwen3-4b":
        proxy: "http://127.0.0.1:7000"
        cmd: |
        /app/llama-server
        -m /models/Qwen3-4B-Instruct-2507-UD-Q4_K_XL.gguf
        # or use `-hf unsloth/Qwen3-4B-Instruct-2507-GGUF:Q4_K_XL` for HuggingFace
        --port 7000

vLLM (Docker)

models:
    "qwen3-4b":
        proxy: "http://127.0.0.1:7000"
        cmd: |
        docker run --name qwen-vllm
        --init --rm -p 7000:8080
        --ipc=host \
        vllm/vllm-openai:latest
        -m /models/Qwen/Qwen3-4B-Instruct-2507
        cmdStop: docker stop qwen-vllm

vLLM (local):

models:
    "qwen3-4b":
        proxy: "http://127.0.0.1:7000"
        cmd: |
        source /app/vllm/.venv/bin/activate && \
        /app/vllm/.venv/bin/vllm serve \
        --port 7000 \
        --host 0.0.0.0 \
        -m /models/Qwen/Qwen3-4B-Instruct-2507
        cmdStop: pkill -f "vllm serve"

Install the container:

We use the cuda tag here, but llama-swap offers cpu, intel, vulkan, and musa tags as well. Releases can be found here.

llama.cpp

docker run -d --gpus all --restart unless-stopped --network app-net --pull=always --name llama-swap -p 9292:8080 \
-v /path/to/models:/models \
-v /home/<your_username>/llama-swap/config.yaml:/app/config.yaml \
-v /home/<your_username>/llama.cpp/build/bin/llama-server:/app/llama-server \
ghcr.io/mostlygeek/llama-swap:cuda

vLLM (Docker/local)

docker run -d --gpus all --restart unless-stopped --network app-net --pull=always --name llama-swap -p 9292:8080 \
-v /path/to/models:/models \
-v /home/<your_username>/vllm:/app/vllm \
-v /home/<your_username>/llama-swap/config.yaml:/app/config.yaml \
ghcr.io/mostlygeek/llama-swap:cuda

Replace <your_username> with your actual username and /path/to/models with the path to your model files.

[!NOTE] llama-swap prefers Docker-based vLLM due to cleanliness of environments and adherence to SIGTERM signals sent by the server. I've written out both options here.

This should result in a functioning llama-swap instance running at http://localhost:9292, which can be confirmed by running curl http://localhost:9292/health. It is highly recommended that you read the configuration documentation. llama-swap is thoroughly documented and highly configurable - utilizing its capabilities will result in a tailored setup ready to deploy as you need it.

`systemd` Service

The other way to persist a model across system reboots is to start the inference engine in a .service file that will run alongside the Linux operating system when booting, ensuring that it is available whenever the server is on. If you're willing to live with the relative compromise of not being able to swap models/backends and are satisfied with running one model, this is the lowest overhead solution and works great.

Let's call the service we're about to build llm-server.service. We'll assume all models are in the models child directory - you can change this as you need to.

Create the systemd service file:

sudo nano /etc/systemd/system/llm-server.service

Configure the service file:

llama.cpp

[Unit]
Description=LLM Server Service
After=network.target

[Service]
User=<user>
Group=<user>
WorkingDirectory=/home/<user>/llama.cpp/build/bin/
ExecStart=/home/<user>/llama.cpp/build/bin/llama-server \
    --port <port> \
    --host 0.0.0.0 \
    -m /home/<user>/llama.cpp/models/<model> \
    --no-webui # [other engine arguments]
Restart=always
RestartSec=10s

[Install]
WantedBy=multi-user.target

vLLM

[Unit]
Description=LLM Server Service
After=network.target

[Service]
User=<user>
Group=<user>
WorkingDirectory=/home/<user>/vllm/
ExecStart=/bin/bash -c 'source .venv/bin/activate && vllm serve --port <port> --host 0.0.0.0 -m /home/<user>/vllm/models/<model>'
Restart=always
RestartSec=10s

[Install]
WantedBy=multi-user.target

Replace <user>, <port>, and <model> with your Linux username, desired port for serving, and desired model respectively.

Reload the systemd daemon:
```
sudo systemctl daemon-reload
```

Run the service:

If llm-server.service doesn't exist:

sudo systemctl enable llm-server.service
sudo systemctl start llm-server

If llm-server.service does exist:

sudo systemctl restart llm-server

(Optional) Check the service's status:
```
sudo systemctl status llm-server
```

Open WebUI Integration

llama-swap

Navigate to Admin Panel > Settings > Connections and set the following values:

Enable OpenAI API
API Base URL: http://llama-swap:8080/v1
API Key: anything-you-like

`systemd` Service

Follow the same steps as above.

Enable OpenAI API
API Base URL: http://localhost:<port>/v1
API Key: anything-you-like

Replace <port> with your desired port.

Chat Platform

Open WebUI

🌟 GitHub
📖 Documentation

Open WebUI is a web-based interface for managing models and chats, and provides a beautiful, performant UI for communicating with your models. You will want to do this if you want to access your models from a web interface. If you're fine with using the command line or want to consume models through a plugin/extension, you can skip this step.

To install without Nvidia GPU support, run the following command:

docker run -d -p 3000:8080 --network app-net --add-host=host.docker.internal:host-gateway -v open-webui:/app/backend/data --name open-webui --restart always ghcr.io/open-webui/open-webui:main

For Nvidia GPUs, run the following command:

docker run -d -p 3000:8080 --network app-net --gpus all --add-host=host.docker.internal:host-gateway -v open-webui:/app/backend/data --name open-webui --restart always ghcr.io/open-webui/open-webui:cuda

You can access it by navigating to http://localhost:3000 in your browser or http://<server_ip>:3000 from another device on the same network. There's no need to add this to the init.bash script as Open WebUI will start automatically at boot via Docker Engine.

MCP Proxy Server

Model Context Protocol (MCP) is a protocol that tools (functions/scripts written in code) to LLMs in a standardized way. Generally, models are being trained more and more with the ability to natively call tools in order to power agentic tasks - think along the lines of having a model use sequential thinking to formulate multiple thoughts, execute multiple targeted web searches, and provide a response leveraging real-time information. MCP also, probably more importantly for most people, enables models to call third-party tools like those for GitHub, Azure, etc. A complete list, curated and maintained by Anthropic, can be found here.

Most guides on the Internet concerning MCP will have you spin up an MCP server via a client like VS Code, Cline, etc. since most agentic uses are for coding or Claude Desktop, which is a proprietary app by Anthropic and not at all what we're aiming to achieve in this guide with respect to privacy. There are other chat clients that support MCP server management from the UI itself (LobeChat, Cherry Studio, etc.), but we want to be able to manage MCP servers in a central and modular way. That way, (i) they aren't tied to a specific client and are available to every client you may use them with and (ii) if you switch chat platforms in the future, your MCP servers require zero changes because they run as a decoupled service - little bit more maintenance for a lot more flexibility down the line. We can do this by setting up an MCP proxy server.

This proxy server will take the MCP servers running via stdio (standard IO) protocol (that can only be accessed by an application running on that device) and make them compatible with streamable HTTP. Any MCP-enabled client can use streamable HTTP so they'll also be able to use all the servers we install on our physical server. This centralizes the management of your MCP servers: create/edit/delete servers in one place, use any of them from your various clients (Open WebUI, VS Code, etc.).

We'll use the fetch, sequential-thinking, and searxng MCP servers to get started. The process for adding more servers will be identical.

mcp-proxy

🌟 GitHub

mcp-proxy is a server proxy that allows switching between transports (stdio to streamable HTTP and vice versa). I'll be using port 3131 to avoid conflicts - feel free to change this as you need to. I will also be extending mcp-proxy to include uv: most MCP servers use either npx or uv and setting mcp-proxy up without uv is going to hamper your ability to run the MCP servers you'd like. If you don't require uv, (i) don't add the build section in the compose file and (ii) skip step 4.

Create a compose file

mkdir mcp-proxy
cd mcp-proxy
sudo nano docker-compose.yaml

Enter the following:

services:
  mcp-proxy:
    container_name: mcp-proxy
    build:
        context: .
        dockerfile: Dockerfile
    networks:
    - app-net
    volumes:
    - .:/config
    - /:/<server_hostname>:ro
    restart: unless-stopped
    ports:
    - 3131:3131
    command: "--pass-environment --port=3131 --host 0.0.0.0 --transport streamablehttp --named-server-config /config/servers.json"

networks:
  app-net:
    external: true

Replace <server_hostname> with your actual server's hostname (or whatever else). This is primarily useful when adding filesystem or similar MCP servers that read from and write files to the file system. Feel free to skip if that isn't your goal.

Create a servers.json file:

{
    "mcpServers": {
        "fetch": {
            "disabled": false,
            "timeout": 60,
            "command": "uvx",
            "args": [
                "mcp-server-fetch"
            ],
            "transportType": "stdio"
        },
        "sequential-thinking": {
            "command": "npx",
            "args": [
                "-y",
                "@modelcontextprotocol/server-sequential-thinking"
            ]
        },
        "searxng": {
            "command": "npx",
            "args": ["-y", "mcp-searxng"],
            "env": {
                "SEARXNG_URL": "http://searxng:8080/search?q=<query>"
            }
        }
    }
}

Create a Dockerfile:

sudo nano Dockerfile

Enter the following:

FROM ghcr.io/sparfenyuk/mcp-proxy:latest

# Install dependencies for nvm and Node.js
RUN apk add --update npm

# Install the 'uv' package
RUN python3 -m ensurepip && pip install --no-cache-dir uv

ENV PATH="/usr/local/bin:/usr/bin:$PATH" \
    UV_PYTHON_PREFERENCE=only-system

ENTRYPOINT ["catatonit", "--", "mcp-proxy"]

Start the container with docker compose up -d

Your mcp-proxy container should be up and running! Adding servers is simple: add the relevant server to servers.json (you can use the same configuration that the MCP server's developer provides for VS Code, it's identical) and then restart the container with docker restart mcp-proxy.

MCPJungle

🌟 GitHub

MCPJungle is another MCP proxy server with a different focus. It focuses on providing more of a "production-grade" experience, a lot of which is disabled by default in the development mode of the application. We'll use the standard development version of the container here on port 4141.

Create a compose file:

mkdir mcpjungle
cd mcpjungle
sudo nano docker-compose.yaml

Enter the following and save:

# MCPJungle Docker Compose configuration for individual users.
# Use this compose file if you want to run MCPJungle locally for your personal MCP management & Gateway.
# The mcpjungle server runs in development mode.
services:
  db:
    image: postgres:latest
    container_name: mcpjungle-db
    environment:
        POSTGRES_USER: mcpjungle
        POSTGRES_PASSWORD: mcpjungle
        POSTGRES_DB: mcpjungle
    ports:
    - "5432:5432"
    networks:
    - app-net
    volumes:
    - db_data:/var/lib/postgresql/data
    healthcheck:
        test: ["CMD-SHELL", "PGPASSWORD=mcpjungle pg_isready -U mcpjungle"]
        interval: 10s
        timeout: 5s
        retries: 5
    restart: unless-stopped

  mcpjungle:
    image: mcpjungle/mcpjungle:${MCPJUNGLE_IMAGE_TAG:-latest-stdio}
    container_name: mcpjungle-server
    environment:
        DATABASE_URL: postgres://mcpjungle:mcpjungle@db:5432/mcpjungle
        SERVER_MODE: ${SERVER_MODE:-development}
        OTEL_ENABLED: ${OTEL_ENABLED:-false}
    ports:
    - "4141:8080"
    networks:
    - app-net
    volumes:
    # Mount host filesystem current directory to enable filesystem MCP server access
    - .:/host/project:ro
    - /home/<your_username>:/host:ro
    # Other options:
    # - ${HOME}:/host/home:ro
    # - /tmp:/host/tmp:rw
    depends_on:
    db:
        condition: service_healthy
    restart: always

volumes:
    db_data:

networks:
  app-net:
    external: true

Start the container with docker compose up -d

Create a tool file:

sudo nano fetch.json

Enter the following and save:

{
    "name": "fetch",
    "transport": "stdio",
    "command": "npx",
    "args": ["mcp-server-fetch"]
}

docker exec -i mcpjungle-server /mcpjungle register -c /host/project/fetch.json

Repeat steps 3 and 4 for every tool mentioned. Commands for sequential-thinking and searxng can be found below.

sequential-thinking

{
    "name": "sequential-thinking",
    "transport": "stdio",
    "command": "npx",
    "args": ["-y", "@modelcontextprotocol/server-sequential-thinking"]
}

searxng

{
    "name": "searxng",
    "transport": "stdio",
    "command": "npx",
    "args": ["-y", "mcp-searxng"],
    "env": {
        "SEARXNG_URL": "http://searxng:8080/search?q=<query>"
    }
}

Comparison

The choice between the two services is yours entirely: I use mcp-proxy because I find the workflow slightly less cumbersome than MCPJungle. Here's a comparison with the stengths of each service.

mcp-proxy > MCPJungle

Servers can just be added to servers.json and will be registered automatically on container restart - MCPJungle requires manual registration of tools via the CLI
Uses the standard MCP syntax that most clients accept for configuration
Lighter footprint in that it doesn't need to spin up a separate database container
Uses stateful connections - MCPJungle spins up a new connection per tool call, which can lead to some performance overhead

MCPJungle > mcp-proxy

Combines all tools under one endpoint, making it very easy to integrate into a chat frontend
Capable of creating a very configurable setup with tool groups, access control, selective tool enabling/disabling
Supports enterprise features like telemetry

Open WebUI Integration

Open WebUI recently added support for streamable HTTP - where once you may have had to use mcpo, Open WebUI's way of automatically generating an OpenAPI-compatible HTTP server, you can use the MCP servers you've set up as-is with no changes.

mcp-proxy

Navigate to Admin Panel > Settings > External Tools. Click the + button to add a new tool and enter the following information:

URL: http://mcp-proxy:<port>/servers/<tool_name>/mcp
API Key: anything-you-like
ID: <tool_name>
Name: <tool_name>

Replace <port> with the port of the MCP service and <tool_name> with the specific tool you're adding.

MCPJungle

Follow the same steps as above. By design, MCPJungle exposes all tools via one endpoint, so you should only have to add once:

URL: http://mcpjungle-server:8080/mcp
API Key: anything-you-like
ID: <tool_name>
Name: <tool_name>

[!IMPORTANT] When configuring models in Open WebUI (via Admin Panel > Settings > Models > my-cool-model > Advanced Params), change the Function Calling parameter from Default to Native. This step will allow the model to use multiple tool calls to formulate a single response instead of just one.

VS Code/Claude Desktop Integration

The steps for integrating your MCP proxy server in another client such as VS Code (Claude Desktop, Zed, etc.) will be similar, if not exactly the same.

Add the following key and value to your mcp.json file:

"your-mcp-proxy-name": {
    "timeout": 60,
    "type": "stdio",
    "command": "npx",
    "args": [
    "mcp-remote",
    "http://<your-server-url>/mcp",
    "--allow-http"
    ]
}

Text-to-Speech Server

Kokoro FastAPI

🌟 GitHub

Kokoro FastAPI is a text-to-speech server that wraps around and provides OpenAI-compatible API inference for Kokoro-82M, a state-of-the-art TTS model. The documentation for this project is fantastic and covers most, if not all, of the use cases for the project itself.

To install Kokoro-FastAPI, run

git clone https://github.com/remsky/Kokoro-FastAPI.git
cd Kokoro-FastAPI
docker compose up --build

The server can be used in two ways: an API and a UI. By default, the API is served on port 8880 and the UI is served on port 7860.

Open WebUI Integration

Navigate to Admin Panel > Settings > Audio and set the following values:

Text-to-Speech Engine: OpenAI
API Base URL: http://host.docker.internal:8880/v1
API Key: anything-you-like
Set Model: kokoro
Response Splitting: None (this is crucial - Kokoro uses a novel audio splitting system)

The server can be used in two ways: an API and a UI. By default, the API is served on port 8880 and the UI is served on port 7860.

Image Generation Server

ComfyUI

🌟 GitHub
📖 Documentation

ComfyUI is a popular open-source graph-based tool for generating images using image generation models such as Stable Diffusion XL, Stable Diffusion 3, and the Flux family of models.

Clone and navigate to the repository:

git clone https://github.com/comfyanonymous/ComfyUI
cd ComfyUI

Set up a new virtual environment:

python3 -m venv comfyui-env
source comfyui-env/bin/activate

Download the platform-specific dependencies:

Nvidia GPUs

pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu121

AMD GPUs

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm6.0

Intel GPUs

Read the installation instructions from ComfyUI's GitHub.

Download the general dependencies:
```
pip install -r requirements.txt
```

Now, we have to download and load a model. Here, we'll use FLUX.1 [dev], a new, state-of-the-art medium-tier model by Black Forest Labs that fits well on an RTX 3090 24GB. Since we want this to be set up as easily as possible, we'll use a complete checkpoint that can be loaded directly into ComfyUI. For a completely customized workflow, CLIPs, VAEs, and models can be downloaded separately. Follow this guide by ComfyUI's creator to install the FLUX.1 models in a fully customizable way.

[!NOTE] FLUX.1 [schnell] HuggingFace (smaller, ideal for <24GB VRAM)

FLUX.1 [dev] HuggingFace (larger, ideal for 24GB VRAM)

Download your desired model into /models/checkpoints.
If you want ComfyUI to be served at boot and effectively run as a service, add the following lines to init.bash:
```
cd /path/to/comfyui
source comfyui/bin/activate
python main.py --listen
```
Replace /path/to/comfyui with the correct relative path to init.bash.

Otherwise, to run it just once, simply execute the above lines in a terminal window.

Open WebUI Integration

Navigate to Admin Panel > Settings > Images and set the following values:

Image Generation Engine: ComfyUI
API Base URL: http://localhost:8188

[!TIP] You'll either need more than 24GB of VRAM or to use a small language model mostly on CPU to use Open WebUI with FLUX.1 [dev]. FLUX.1 [schnell] and a small language model, however, should fit cleanly in 24GB of VRAM, making for a faster experience if you intend to regularly use both text and image generation together.

SSH

Enabling SSH allows you to connect to the server remotely. After configuring SSH, you can connect to the server from another device on the same network using an SSH client like PuTTY or the terminal. This lets you run your server headlessly without needing a monitor, keyboard, or mouse after the initial setup.

On the server:

Run the following command:
```
sudo apt install openssh-server
```
Start the SSH service:
```
sudo systemctl start ssh
```
Enable the SSH service to start at boot:
```
sudo systemctl enable ssh
```
Find the server's IP address:
```
ip a
```

On the client:

Connect to the server using SSH:
```
ssh <username>@<ip_address>
```
Replace <username> with your username and <ip_address> with the server's IP address.

[!NOTE] If you expect to tunnel into your server often, I highly recommend following this guide to enable passwordless SSH using ssh-keygen and ssh-copy-id. It worked perfectly on my Debian system despite having been written for Raspberry Pi OS.

Firewall

Setting up a firewall is essential for securing your server. The Uncomplicated Firewall (UFW) is a simple and easy-to-use firewall for Linux. You can use UFW to allow or deny incoming and outgoing traffic to and from your server.

Install UFW:
```
sudo apt install ufw
```
Allow SSH, HTTPS, and HTTP to your local network:
```
# Allow all <ip_range> hosts access to port <port>
sudo ufw allow from <ip_range> to any port <port> proto tcp
```
Start with running the above command to open ports 22 (SSH), 80 (HTTP), and 443 (HTTPS) to our local network. Since we use the app-net Docker network for our containers, there's no need to open anything else up. Open ports up carefully, ideally only to specific IPs or to your local network. To allow a port for a specific IP, you can replace the IP range with a single IP and it'll work exactly the same way.

[!TIP] You can find your local network's IP address range by running ip route show. The result will be something like this:
me@my-cool-server:~$ ip route show
default via <router_ip> dev enp3s0 proto dhcp src <server_ip> metric 100
<network_ip_range> dev enp3s0 proto kernel scope link src <server_ip> metric 100
# more routes

Enable UFW:
```
sudo ufw enable
```
Check the status of UFW:
```
sudo ufw status verbose
```

[!WARNING] Enabling UFW without allowing access to port 22 will disrupt your existing SSH connections. If you run a headless setup, this means connecting a monitor to your server and then allowing SSH access through UFW. Be careful to ensure that this port is allowed when making changes to UFW's configuration.

Refer to this guide for more information on setting up UFW.

Remote Access

Remote access refers to the ability to access your server outside of your home network. For example, when you leave the house, you aren't going to be able to access http://<your_server_ip>, because your network has changed from your home network to some other network (either your mobile carrier's or a local network in some other place). This means that you won't be able to access the services running on your server. There are many solutions on the web that solve this problem and we'll explore some of the easiest-to-use here.

Tailscale

Tailscale is a peer-to-peer VPN service that combines many services into one. Its most common use-case is to bind many different devices of many different kinds (Windows, Linux, macOS, iOS, Android, etc.) on one virtual network. This way, all these devices can be connected to different networks but still be able to communicate with each other as if they were all on the same local network. Tailscale is not completely open source (its GUI is proprietary), but it is based on the Wireguard VPN protocol and the remainder of the actual service is open source. Comprehensive documentation on the service can be found here and goes into many topics not mentioned here - I would recommend reading it to get the most of out the service.

On Tailscale, networks are referred to as tailnets. Creating and managing tailnets requires creating an account with Tailscale (an expected scenario with a VPN service) but connections are peer-to-peer and happen without any routing to Tailscale servers. This connection being based on Wireguard means 100% of your traffic is encrypted and cannot be accessed by anyone but the devices on your tailnet.

Installation

First, create a tailnet through the Admin Console on Tailscale. Download the Tailscale app on any client you want to access your tailnet from. For Windows, macOS, iOS, and Android, the apps can be found on their respective OS app stores. After signing in, your device will be added to the tailnet.

For Linux, the steps required are as follows.

Install Tailscale

curl -fsSL https://tailscale.com/install.sh | sh

Start the service
```
sudo tailscale up
```

For SSH, run sudo tailscale up --ssh.

Exit Nodes

An exit node allows access to a different network while still being on your tailnet. For example, you can use this to allow a server on your network to act as a tunnel for other devices. This way, you can not only access that device (by virtue of your tailnet) but also all the devices on the host network its on. This is useful to access non-Tailscale devices on a network.

To advertise a device on as an exit node, run sudo tailscale up --advertise-exit-node. To allow access to the local network via this device, add the --exit-node-allow-lan-access flag.

Local DNS

If one of the devices on your tailnet runs a DNS-sinkhole service like Pi-hole, you'll probably want other devices to use it as their DNS server. Assume this device is named poplar. This means every networking request made by a any device on your tailnet will send this request to poplar, which will in turn decide whether that request will be answered or rejected according to your Pi-hole configuration. However, since poplar is also one of the devices on your tailnet, it will send networking requests to itself in accordance with this rule and not to somewhere that will actually resolve the request. Thus, we don't want such devices to accept the DNS settings according to the tailnet but follow their otherwise preconfigured rules.

To reject the tailnet's DNS settings, run sudo tailscale up --accept-dns=false.

Third-Party VPN Integration

Tailscale offers a Mullvad VPN exit node add-on with their service. This add-on allows for a traditional VPN experience that will route your requests through a proxy server in some other location, effectively masking your IP and allowing the circumvention of geolocation restrictions on web services. Assigned devices can be configured from the Admin Console. Mullvad VPN has proven their no-log policy and offers a fixed $5/month price no matter what duration you choose to pay for.

To use a Mullvad exit on one of your devices, first find the exit node you want to use by running sudo tailscale exit-node list. Note the IP and run sudo tailscale up --exit-node=<your_chosen_exit_node_ip>.

[!WARNING] Ensure the device is allowed to use the Mullvad add-on through the Admin Console first.

Updating

Updating your system is a good idea to keep software running optimally and with the latest security patches. Updates to Ollama allow for inference from new model architectures and updates to Open WebUI enable new features like voice calling, function calling, pipelines, and more.

I've compiled steps to update these "primary function" installations in a standalone section because I think it'd be easier to come back to one section instead of hunting for update instructions in multiple subsections.

General

Upgrade Debian packages by running the following commands:

sudo apt update
sudo apt upgrade

Nvidia Drivers & CUDA

Follow Nvidia's guide here to install the latest CUDA drivers.

[!WARNING] Don't skip this step. Not installing the latest drivers after upgrading Debian packages will throw your installations out of sync, leading to broken functionality. When updating, target everything important at once. Also, rebooting after this step is a good idea to ensure that your system is operating as expected after upgrading these crucial drivers.

Ollama

Rerun the command that installs Ollama - it acts as an updater too:

curl -fsSL https://ollama.com/install.sh | sh

llama.cpp

Enter your llama.cpp folder and run the following commands:

cd llama.cpp
git pull
# Rebuild according to your setup - uncomment `-DGGML_CUDA=ON` for CUDA support
cmake -B build # -DGGML_CUDA=ON
cmake --build build --config Release

vLLM

For a manual installation, enter your virtual environment and update via pip:

source vllm/.venv/bin/activate
pip install vllm --upgrade

For a Docker installation, you're good to go when you re-run your Docker command, because it pulls the latest Docker image for vLLM.

llama-swap

Delete the current container:

docker stop llama-swap
docker rm llama-swap

Re-run the container command from the llama-swap section.

Open WebUI

To update Open WebUI once, run the following command:

docker run --rm --volume /var/run/docker.sock:/var/run/docker.sock containrrr/watchtower --run-once open-webui

To keep it updated automatically, run the following command:

docker run -d --name watchtower --volume /var/run/docker.sock:/var/run/docker.sock containrrr/watchtower open-webui

mcp-proxy/MCPJungle

Navigate to the directory and pull the latest container image:

cd mcp-proxy # or mcpjungle
docker compose down
docker compose pull
docker compose up -d

Kokoro FastAPI

Navigate to the directory and pull the latest container image:

cd Kokoro-FastAPI
docker compose pull
docker compose up -d

ComfyUI

Navigate to the directory, pull the latest changes, and update dependencies:

cd ComfyUI
git pull
source comfyui-env/bin/activate
pip install -r requirements.txt

Troubleshooting

Docker

Since we aim to use our user instead of root to run services, you may encounter permissions issues when trying to mount a volume that is not technically owned by our user. To fix this, you can do one of two things:
- Change the ownership of the directory:
```
sudo chown -R $(id -u):$(id -g) /path/to/volume
```
- Grant access to our user with access control lists (ACLs, in short):
```
# Grant read/write to specific user
sudo setfacl -R -m u:$(id -u):rw /path/to/volume

# Grant to group
sudo setfacl -R -m g:$(id -g):rw /path/to/volume
```
  In case you don't have the acl package, install with sudo apt update && sudo apt install acl. The official package page can be found here.
  
  Replace /path/to/volume with the desired path.
  
  I like ACLs better - it solves the access issue cleanly without changing ownership unnecessarily and, in my experience, tends to break things less. However, if the resource is clearly meant to be owned by a specific user, changing ownership is preferred purely for separation of concerns. Gauge for yourself, both are completely acceptable options. Avoid changing permissions if your container is already up and running, though.

`ssh`

If you encounter an issue using ssh-copy-id to set up passwordless SSH, try running ssh-keygen -t rsa on the client before running ssh-copy-id. This generates the RSA key pair that ssh-copy-id needs to copy to the server.

Nvidia Drivers

Disable Secure Boot in the BIOS if you're having trouble with the Nvidia drivers not working. For me, all packages were at the latest versions and nvidia-detect was able to find my GPU correctly, but nvidia-smi kept returning the NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver error. Disabling Secure Boot fixed this for me. Better practice than disabling Secure Boot is to sign the Nvidia drivers yourself but I didn't want to go through that process for a non-critical server that can afford to have Secure Boot disabled.
If you run into docker: Error response from daemon: unknown or invalid runtime name: nvidia., you probably have --runtime nvidia in your Docker statement. This is meant for nvidia-docker, which is deprecated now. Removing this flag from your command should get rid of this error.

Ollama

If you receive the could not connect to ollama app, is it running? error, your Ollama instance wasn't served properly. This could be because of a manual installation or the desire to use it at-will and not as a service. To run the Ollama server once, run:
```
ollama serve
```
Then, in a new terminal, you should be able to access your models regularly by running:
```
ollama run <model>
```
For detailed instructions on manually configuring Ollama to run as a service (to run automatically at boot), read the official documentation here. You shouldn't need to do this unless your system faces restrictions using Ollama's automated installer.
If you receive the Failed to open "/etc/systemd/system/ollama.service.d/.#override.confb927ee3c846beff8": Permission denied error from Ollama after running systemctl edit ollama.service, simply creating the file works to eliminate it. Use the following steps to edit the file.
- Run:
```
sudo mkdir -p /etc/systemd/system/ollama.service.d
sudo nano /etc/systemd/system/ollama.service.d/override.conf
```
- Retry the remaining steps.
If you still can't connect to your API endpoint, check your firewall settings. This guide to UFW (Uncomplicated Firewall) on Debian is a good resource.

vLLM

If you encounter RuntimeError: An error occurred while downloading using `hf_transfer`. Consider disabling HF_HUB_ENABLE_HF_TRANSFER for better error handling., add HF_HUB_ENABLE_HF_TRANSFER=0 to the --env flag after your HuggingFace Hub token. If this still doesn't fix the issue -
- Ensure your user has all the requisite permissions for HuggingFace to be able to write to the cache. To give read+write access over the HF cache to your user (and, thus, huggingface-cli), run:
```
sudo chmod 777 ~/.cache/huggingface
sudo chmod 777 ~/.cache/huggingface/hub
```
- Manually download a model via the HuggingFace CLI and specify --download-dir=~/.cache/huggingface/hub in the engine arguments. If your .cache/huggingface directory is being troublesome, specify another directory to the --download-dir in the engine arguments and remember to do the same with the --local-dir flag in any huggingface-cli commands.

Open WebUI

If you encounter Ollama: llama runner process has terminated: signal: killed, check your Advanced Parameters, under Settings > General > Advanced Parameters. For me, bumping the context length past what certain models could handle was breaking the Ollama server. Leave it to the default (or higher, but make sure it's still under the limit for the model you're using) to fix this issue.

Monitoring

To monitor GPU usage, power draw, and temperature, you can use the nvidia-smi command. To monitor GPU usage, run:

watch -n 1 nvidia-smi

This will update the GPU usage every second without cluttering the terminal environment. Press Ctrl+C to exit.

Notes

This is my first foray into setting up a server and ever working with Linux so there may be better ways to do some of the steps. I will update this repository as I learn more.

Software

I chose Debian because it is, apparently, one of the most stable Linux distros. I also went with an XFCE desktop environment because it is lightweight and I wasn't yet comfortable going full command line.
Use a user for auto-login, don't log in as root unless for a specific reason.
To switch to root in the command line without switching users, run sudo -i.
If something using a Docker container doesn't work, try running docker ps -a to see if the container is running. If it isn't, try running docker compose up -d again. If it is and isn't working, try running docker restart <container_id> to restart the container.
If something isn't working no matter what you do, try rebooting the server. It's a common solution to many problems. Try this before spending hours troubleshooting. Sigh.
While it takes some time to get comfortable with, using an inference engine like llama.cpp and vLLM (as compared to Ollama) is really the way to go to squeeze the maximum performance out of your hardware. If you're reading this guide in the first place and haven't already thrown up your hands and used a cloud provider, it's a safe assumption that you care about the ethos of hosting all this stuff locally. Thus, get your experience as close to a cloud provider as it can be by optimizing your server.

Hardware

The power draw of my EVGA FTW3 Ultra RTX 3090 was 350W at stock settings. I set the power limit to 250W and the performance decrease was negligible for my use case, which is primarily code completion in VS Code and the Q&A via chat.
Using a power monitor, I measured the power draw of my server for multiple days - the running average is ~60W. The power can spike to 350W during prompt processing and token generation, but this only lasts for a few seconds. For the remainder of the generation time, it tended to stay at the 250W power limit and dropped back to the average power draw after the model wasn't in use for about 20 seconds.
Ensure your power supply has enough headroom for transient spikes (particularly in multi GPU setups) or you may face random shutdowns. Your GPU can blow past its rated power draw and also any software limit you set for it based on the chip's actual draw. I usually aim for +50% of my setup's estimated total power draw.

References

Adding user to sudo group:

https://askubuntu.com/questions/168280/how-do-i-grant-sudo-privileges-to-an-existing-user

Downloading Nvidia drivers:

Downloading AMD drivers:

https://wiki.debian.org/AtiHowTo

Secure Boot:

https://askubuntu.com/a/927470

Monitoring GPU usage, power draw:

https://unix.stackexchange.com/questions/38560/gpu-usage-monitoring-cuda/78203#78203

Passwordless sudo:

Auto-login:

Expose Ollama to LAN:

Firewall:

https://www.digitalocean.com/community/tutorials/how-to-set-up-a-firewall-with-ufw-on-debian-10

Passwordless ssh:

https://www.raspberrypi.com/documentation/computers/remote-access.html#configure-ssh-without-a-password

Adding CUDA to PATH:

https://askubuntu.com/questions/885610/nvcc-version-command-says-nvcc-is-not-installed

Docs:

Acknowledgements

Cheers to all the fantastic work done by the open-source community. This guide wouldn't exist without the effort of the many contributors to the projects and guides referenced here. To stay up-to-date on the latest developments in the field of machine learning, LLMs, and other vision/speech models, check out r/LocalLLaMA. To stay in touch with (or fall down the rabbit hole of) the world of self-hosted apps out there for your new server, check out r/selfhosted.

[!NOTE] Please star any projects you find useful and consider contributing to them if you can. Stars on this guide would also be appreciated if you found it helpful, as it helps others find it too.

LLM Server Docs

Reviews

Documentation

Local LLaMA Server Setup Documentation

Table of Contents

About

Priorities

Prerequisites

General

Allow sudo Permissions

Update System Packages

Schedule Startup Script

Configure Script Permissions

Configure Auto-Login (optional)

Docker

Add User to Docker Group

Nvidia Container Toolkit

Create a Network

Harden Docker Containers

Helpful Commands

HuggingFace CLI

Manage Models

Download Models

Delete Models

Search Engine

SearXNG

Open WebUI Integration

Inference Engine

Ollama

llama.cpp

vLLM

Open WebUI Integration

Ollama vs. llama.cpp

vLLM vs. Ollama/llama.cpp

Model Server

llama-swap

systemd Service

Open WebUI Integration

llama-swap

systemd Service

Chat Platform

Open WebUI

MCP Proxy Server

mcp-proxy

MCPJungle

Comparison

Open WebUI Integration

mcp-proxy

MCPJungle

VS Code/Claude Desktop Integration

Text-to-Speech Server

Kokoro FastAPI

Open WebUI Integration

Image Generation Server

ComfyUI

Open WebUI Integration

SSH

Firewall

Remote Access

Tailscale

Installation

Exit Nodes

Local DNS

Third-Party VPN Integration

Updating

General

Nvidia Drivers & CUDA

Ollama

llama.cpp

vLLM

llama-swap

Open WebUI

mcp-proxy/MCPJungle

Kokoro FastAPI

ComfyUI

Troubleshooting

Docker

ssh

Nvidia Drivers

Ollama

Allow `sudo` Permissions

`systemd` Service

`systemd` Service

`ssh`