CodeProject.AI Server: AI the easy way.

CodeProject

Rate me:

5.00/5 (95 votes)

27 May 202416 min read

5.1M

577.7K

280

7.3K

Version 2.6.5. Our fast, free, self-hosted Artificial Intelligence Server for any platform, any language

CodeProject.AI Server is a locally installed, self-hosted, fast, free and Open Source Artificial Intelligence server for any platform, any language. No off-device or out of network data transfer, no messing around with dependencies, and able to be used from any platform, any language. Runs as a Windows Service or a Docker container.

Pre-release

Quick Links

CodeProject.AI Server: An Artificial Intelligence Server

For those who want to integrate AI functionality into their applications without writing the AI functionality or dealing with the insanely painful task of ensuring everything is setup correctly. CodeProject.AI Server manages your MLOps for you.

Think of CodeProject.AI Server like a database server: you install it, it runs in the background, and provides AI operations for any application via a simple API. The AI operations are handled by drop-in modules that can be easily created using any language, any stack, as long as that stack runs on the host machine. Python, .NET, node - whatever works for you.

CodeProject.AI server runs as a Windows service, under systemd in Linux, or on startup on macOS. Alternatively there are multiple Docker images for x64, arm64 and CUDA enabled systems. Any language that can make HTTP calls can access the service, and the server does not require an external internet connection. Your data stays in your network.


Windows	macOS	macOS-arm64	Ubuntu	Debian	Raspberry Pi	Orange Pi	Jetson Nano	Docker

What Does It Do?

The CodeProject.AI Server's Dashboard

Currently CodeProject.AI Server contains AI modules that provide:

Object Detection (Python and .NET versions that use YOLO, plus a Tensorflow-Lite module that's ultra-lightweight and great for Raspberry Pi and Coral USB sticks
Face Detection and recognition
Text processing such as sentiment analysis and summarization
Image processing such as background removal, background blur, cartoon-isation and resolution enhancement
Model training, including dataset acquisition, for YOLO object detection

How Do I Use It?

Install the server and start making calls to the API. It's that easy.

Guides, Help, FAQs

CodeProject.AI Server	Home Assistant	Blue Iris
Installing on Windows Installing on Linux Installing on macOS Running in Docker Setting up the Dev environment Common Issues	Setup on a Raspberry Pi Setting up CodeProject.AI Server on Home Assistant OS Common Install Issues	Installing Blue Iris Blue Iris and CodeProject.AI Server Common Issues with Blue Iris and CodeProject.AI Server

The CodeProject.AI Server's Explorer in action

Why We Built CodeProject.AI Server

AI programming is something every single developer should be aware of
We wanted a fun project we could use to help teach developers and get them involved in AI. We'll be using CodeProject.AI Server as a focus for articles and exploration to make it fun and painless to learn AI programming.
We want your contributions!
AI coding examples have too many moving parts
You need to install packages and languages and extensions to tools, and then updates and libraries (but version X, not version Y) and then you have to configure paths and...Oh, you want to run on Windows not Linux? In that case, you need to... It's all too hard. There was much yelling at CodeProject.
CodeProject.AI Server includes everything you need in a single installer. CodeProject.AI Server also provides an installation script that will setup your dev environment and get you debugging within a couple of clicks.
AI solutions often require the use of cloud services
If you trust the cloud provider, or understand the billing structure, or can be assured you aren't sending sensitive data or won't go over the free tier, this is fine. If you have a webcam inside your house, or can't work out how much AWS will charge, it's not so OK.
CodeProject.AI Server can be installed locally. Your machine, your network, no data needs to leave your device.

1: Running and Playing With the Features

Install and Run
1. For a Windows Service, download the latest version, install, and launch the shortcut to the server's dashboard on your desktop or open a browser to http://localhost:32168.
  If you wish to take advantage of a CUDA enabled NVIDIA GPU, please ensure you have the CUDA drivers installed before you install CodeProject.AI. We recommend CUDA 11.8 if running Windows
2. For a Docker Container for 64 Bit Linux, run:
```
docker run -p 32168:32168 --name CodeProject.AI -d codeproject/ai-server
```
  For Docker GPU (supports NVIDIA CUDA), please use:
```
docker run --gpus all -p 32168:32168 --name CodeProject.AI -d codeproject/ai-server:cuda11_7
```
On the dashboard, at the top, is a link to the demo playground. Open that and play!

2: Running and Debugging the Code

Clone the CodeProject CodeProject.AI Server repository.
Make sure you have Visual Studio Code or Visual Studio 2019+ installed.
Run the setup script in /src
Debug the front-end server application (see notes below, but it's easy).

3. Using CodeProject.AI Server in My Application

Here's an example of using the API for scene detection using a simple JavaScript call:

HTML

<html>
<body>
Detect the scene in this file: <input id="image" type="file" />
<input type="button" value="Detect Scene" onclick="detectScene(image)" />

<script>
function detectScene(fileChooser) {
    var formData = new FormData();
    formData.append('image', fileChooser.files[0]);

    fetch('http://localhost:5000/v1/vision/detect/scene', {
        method: "POST",
        body: formData
    })
    .then(response => {
        if (response.ok) response.json().then(data => {
            console.log(`Scene is ${data.label}, ${data.confidence} confidence`)
        });
    });
}
</script>
</body>
</html>

You can include the CodeProject.AI Server installer (or just a link to the latest version of the installer) in your own apps and installers and voila, you have an AI enabled app.

See the API documentation for a complete rundown of functionality.

Notes on the installers

The native installers (Windows, Ubuntu and macOS) all install the server as a service. On Windows it's a Windows service, on Ubuntu it uses systemd, and on macOS it's simply a login item so will start each time you login.

For all platforms, open http://localhost:32168 to view the dashboard.

To uninstall, please take note of the instructions when you install. For reference:

Windows uses the standard Windows installer, so use the Control Panel / Apps and Features applet to manage the installation.
Ubuntu uses dpkg, so to uninstall simply call
Bash
```
sudo dpkg -r codeproject.ai-server
```

macOS uninstall is via the command line

Shell

sudo bash "/Library/CodeProject.AI Server/<version>/uninstall.sh"

Notes on CUDA and Nvidia Support

If you have a CUDA enabled Nvidia card, please then ensure you

install the CUDA Drivers (We recommend CUDA 11.7 or CUDA 11.8 if running Windows)
Install CUDA Toolkit 11.8.
Download and run our cuDNN install script to install cuDNN 8.9.4.

Nvidia downloads and drivers are challenging! Please ensure you download a driver that is compatible with CUDA 11.7+, which generally means the CUDA driver version 516.94 or below. Version 522.x or above may not work. You may need to refer to the release notes for each driver to confirm.

Our Docker images are based on CUDA 11.7 (for legacy reasons) and 12.2. As long as you have a driver installed that can handle 11.7 or 12.2 then the docker image will interface with your drivers and work fine.

CUDA 12.2 brings a few challenges with code that uses PyTorch due to the move to Torch 2.0, so we tend to favour 11.7. Some older cards will not be compatible with CUDA 12, or even CUDA 11.7. If you are struggling with older cards that don't support CUDA 11.7 then post a comment and we'll try and help.

Since we are using CUDA 11.7+ (which has support for compute capability 3.7 and above), we can only support Nvidia CUDA cards that are equal to or better than a GK210 or Tesla K80 card. Please refer to this table of supported cards to determine if your card has compute capability 3.7 or above.

Newer cards such as the GTX 10xx, 20xx and 30xx series, RTX, MX series are fully supported.

AI is a memory intensive operation. Some cards with 2GB RAM or less may struggle in some situations. Using the dashboard, you can either disable modules you don't need, or disable GPU support entirely for one or more modules. This will free up memory and help get you back on track.

What Does It Include?

CodeProject.AI Server includes:

A HTTP REST API Server. The server listens for requests from other apps, passes them to the backend analysis services for processing, and then passes the results back to the caller. It runs as a simple self-contained web service on your device.
Backend Analysis services. The brains of the operation is in the analysis services sitting behind the front end API. All processing of data is done on the current machine. No calls to the cloud and no data leaving the device.
The source code, naturally.

CodeProject.AI Server can currently

Detect objects in images
Detect faces in images
Detect the type of scene represented in an image
Recognise faces that have been registered with the service
Perform detection on custom models

The development environment also provides modules that can

Remove a background from an image
Blur a background from an image
Enhance the resolution of an image
Pull out the most important sentences in text to generate a text summary
Prove sentiment analysis on text

We will be constantly expanding the feature list.

Our Goals

To promote AI development and inspire the AI developer community to dive in and have a go. Artificial Intelligence is a huge paradigm change in the industry and all developers owe it to themselves to experiment in and familiarize themselves with the technology. CodeProject.AI Server was built as a learning tool, a demonstration, and a library and service that can be used out of the box.
To make AI development easy. It's not that AI development is that hard. It's that there are so, so many options. Our architecture is designed to allow any AI implementation to find a home in our system, and for our service to be callable from any language.
To focus on core use-cases. We're deliberately not a solution for everyone. Instead, we're a solution for common day-to-day needs. We will be adding dozens of modules and scores of AI capabilities to our system, but our goal is always clarity and simplicity over a 100% solution.
To tap the expertise of the Developer Community. We're not experts but we know a developer or two out there who are. The true power of CodeProject.AI Server comes from the contributions and improvements from our AI community.

License

CodeProject.AI Server is licensed under the Server-Side Public License.

Release Notes

What's New - 2.6

You can now select, at install time, which modules you wish to have initially installed
Some modules (Coral, Yolov8) now allow you to download individual models at runtime via the dashboard.
A new generative AI module (Llama LLM Chatbot)
A standardised way to handle (in code) modules that run long processes such as generative AI
Debian support has been improved
Small UI improvements to the dashboard
Some simplification of the modulesettings files
The inclusion, in the source code, of template .NET and Python modules (both simple and long process demos)
Improvements to the Coral and ALPR modules (thanks to Seth and Mike)
Docker CUDA 12.2 image now includes cuDNN
Install script fixes
Added Object Segmentation to the YOLOv8 module
2.6.5 Various installer fixes

Previous Versions

Release 2.5

Dynamic Explorer UI: Each module now supplies its own UI for the explorer
Improved dashboard and explorer
- The module listing now shows module version history if you click the version number
- Explorer benchmark has been updated to use the custom models of the currently active object detection module
- The Info button on the dashboard now includes a status data dump from the module. For things like object detectors, it will include a dictionary of labels / counts so you can see what's being detected. For longer running modules such as training it will include the training status. This is here to enable better UI features in the future
Updated module settings schema that includes module author and original project acknowledgement
Installer fixes
Improved Jetson support
Lots of bug fixes, but specifically there was a script issue affecting module installs, and a modulesettings.json issue affecting the YOLOv5 6.2 module, as well as the SuperResolution module.
Updated ALPR, OCR (PP-OCR4 support thanks to Mike Lud) and Coral Object Detection (multi-TPU support thanks to Seth Price) modules
Pre-installed modules in Docker can now be uninstalled / reinstalled
A new Sound Classifier module has been included
2.5.4: A separate status update from each module that decouples the stats for a module. This just cleans things up a little on the backend
2.5.4: Minor modulesettings.json schema update, which introduces the concept of model requirements.
2.5.5: Support for long running processes with accompanying stable difussion module.

Release 2.4

Mesh support Automatically offload inference work to other servers on your network based on inference speed. Zero config, and dashboard support to enable/disable.
CUDA detection fixed
Module self-test performed on installation
YOLOv8 module added
YOLOv5 .NET module fixes for GPU, and YOLOv5 3.1 GPU support fixed
Python package and .NET installation issues fixed
Better prompts for admin-only installs
More logging output to help diagnose issues
VC Redist hash error fixed
General bug fixes.
Breaking: modulesettings.json schema changed

Release 2.3

A focus on improving the installation of modules at runtime. More error checks, faster re-install, better reporting, and manual fallbacks in situations where admin rights are needed
A revamped SDK that removes much (or all, in some cases) of the boilerplate code needed in install scripts
Fine grained support for different CUDA versions as well as systems such as Raspberry Pi, Orange Pi and Jetson
Support for CUDA 12.2
GPU support for PaddlePaddle (OCR and license plate readers benefit)
CUDA 12.2 Docker image
Lots of bug fixes in install scripts
UI tweaks
2.3.4 ALPR now using GPU in Windows
2.3.4 Corrections to Linux/macOS installers

Release 2.2.0

This release is still in testing and is focussed mainly on the installation process

An entirely new Windows installer offering more installation options and a smoother upgrade experience from here on.
New macOS and Ubuntu native installers, for x64 and arm64 (including Raspberry Pi)
A new installation SDK for making module installers far easier
Improved installation feedback and self-checks
Coral.AI support for Linux, macOS (version 11 and 12 only) and Windows
Updates:
- 2.2.1 - 2.2.3 various installer fixes
- 2.2.4 - Fix to remove chunking in order to allow HTTP1.1 access to the API (Blue Iris fix)

Release 2.1.x Beta

Improved Raspberry Pi support. A new, fast object detection module with support for the Coral.AI TPU, all within an Arm64 Docker image
All modules can now be installed / uninstalled (rather than having some modules fixed and uninstallable).
Installer is streamlined: Only the server is installed at installation time, and on first run, we install Object Detection (Python and .NET) and Face Processing (which can be uninstalled).
Reworking of the Python module SDK. Modules are new child classes, not aggregators of our module runner.
Reworking of the modulesettings file to make it simpler and have less replication
Improved logging: quantity, quality, filtering and better information
Addition of two modules: ObjectDetectionTFLite for Object Detection on Raspberry Pi using Coral, and Cartoonise for some fun
Improvements to half-precision support checks on CUDA cards
Modules are now versioned and our module registry will now only show modules that fit your current server version.
Various bug fixes
Shared Python runtimes now in /runtimes.
All modules moved from the /AnalysisLayer folder to the /modules folder
Tested on CUDA 12
Patch 2.1.11: YOLO training modulke now allows you to use your own dataset. YOLO 6.2 / Face Processing reverted back to Torch 1.13.
Patch 2.1.10: Added YOLOv5 training module and support. Improved system info. Orange Pi and NVIDIA Jetson support. Added Triggers. Renamed VersionCompatibililty to ModuleReleases. Becoz speling.
Patch 2.1.9: Increased and adjustable module install timeout and improved install logs. Fixes around resource contention in PyTorch, Fixes to resource usage reporting, improved Native Linux/WSL CUDA setup. Async fixes. Improvements to half-precision support.
Patch 2.1.8: Reduced, drastically, the load on the system while getting CPU/GPU usage updates.
Patch 2.1.7: Fixed a memory / resource leak that may have been causing server shutdowns
Patch 2.1.6 and below: Installer fixes

Please see our CUDA Notes for information on setting up, and restrictions around, Nvidia cards and CUDA support.

If you are upgrading: when the dashboard launches, it might be necessary to force-reload (Ctrl+R on Windows) the dashboard to ensure you are viewing the latest version.

Release 2.0.x Beta

2.0.8: Improved analysis process management. Stamp out those errant memory hogging Python processes!
2.0.7: Improved logging, both file based and in the dashboard, module installer/uninstaller bug fixes
2.0.6: Corrected issues with downloadable modules installer
Our new Module Registry: download and install modules at runtime via the dashboard
Improved performance for the Object Detection modules
Optional YOLO 3.1 Object Detection module for older GPUs
Optimised RAM use
Support for Raspberry Pi 4+. Code and run natively directly on the Raspberry Pi using VSCode natively
Revamped dashboard
New timing reporting for each API call
New, simplified setup and install scripts

Release 1.6.x Beta

Optimised RAM use
Ability to enable / disable modules and GPU support via the dashboard
REST settings API for updating settings on the fly
Apple M1/M2 GPU support
Workarounds for some Nvidia cards
Async processes and logging for a performance boost
Breaking: The CustomObjectDetection is now part of ObjectDetectionYolo
Performance fix for CPU + video demo
Patch 1.6.7: potential memory leak addressed
Patch 1.6.8: image handling improvements on Linux, multi-thread ONNX on .NET

Release 1.5.6.2 Beta

Docker nVidia GPU support
Further performance improvements
cuDNN install script to help with nVidia driver and toolkit installation
Bug fixes

Release 1.5.6 Beta

nVidia GPU support for Windows
Perf improvements to Python modules
Work on the Python SDK to make creating modules easier
Dev installers now drastically simplified for those creating new modules
Added SuperResolution as a demo module

Release 1.5 Beta

Support for custom models

Release 1.3.x Beta

Refactored and improved setup and module addition system
Introduction of modulesettings.json files
New analysis modules

Release 1.2.x Beta

Support for Apple Silicon for development mode
Native Windows installer
Runs as Windows Service
Run in a Docker Container
Installs and builds using VSCode in Linux (Ubuntu), macOS and Windows, as well as Visual Studio on Windows
General optimisation of the download payload sizes

We started with a proof of concept on Windows 10+ only. Installs we via a simple BAT script, and the code is full of exciting sharp edges. A simple dashboard and playground are included. Analysis is currently Python code only.
Version checks are enabled to alert users to new versions.
A new .NET implementation scene detection using the YOLO model to ensure the codebase is platform and tech stack agnostic
Blue Iris integration completed.

Written By

CodeProject

Software Developer CodeProject Solutions

Canada

The CodeProject team have been writing software, building communities, and hosting CodeProject.com for over 20 years. We are passionate about helping developers share knowledge, learn new skills, and connect. We believe everyone can code, and every contribution, no matter how small, helps.

The CodeProject team is currently focussing on CodeProject.AI Server, a stand-alone, self-hosted server that provides AI inferencing services on any platform for any language. Learn AI by jumping in the deep end with us: codeproject.com/AI.

This is a Organisation

4 members

Server version: 2.6.5 System: Windows Operating System: Windows (Microsoft Windows 10.0.19045) CPUs: Intel(R) Core(TM) i5-7500 CPU @ 3.40GHz (Intel) 1 CPU x 4 cores. 4 logical processors (x64) GPU (Primary): Intel(R) HD Graphics 630 (1,024 MiB) (Intel Corporation) Driver: 31.0.101.2111 System RAM: 16 GiB Platform: Windows BuildConfig: Release Execution Env: Native Runtime Env: Production Runtimes installed: .NET runtime: 7.0.20 .NET SDK: 7.0.410 Default Python: Not found Go: Not found NodeJS: Not found Rust: Not found Video adapter info: Intel(R) HD Graphics 630: Driver Version 31.0.101.2111 Video Processor Intel(R) HD Graphics Family System GPU info: GPU 3D Usage 41% GPU RAM Usage 0 Global Environment variables: CPAI_APPROOTPATH = <root> CPAI_PORT = 32168

Module 'Object Detection (YOLOv5 .NET)' 1.10.2 (ID: ObjectDetectionYOLOv5Net) Valid: True Module Path: <root>\modules\ObjectDetectionYOLOv5Net Module Location: Internal AutoStart: True Queue: objectdetection_queue Runtime: dotnet Runtime Location: System FilePath: bin\ObjectDetectionYOLOv5Net.exe Start pause: 1 sec Parallelism: 0 LogVerbosity: Platforms: all,!windows-arm64 GPU Libraries: installed if available GPU: use if supported Accelerator: Half Precision: enable Environment Variables CUSTOM_MODELS_DIR = <root>\modules\ObjectDetectionYOLOv5Net\custom-models MODELS_DIR = <root>\modules\ObjectDetectionYOLOv5Net\assets MODEL_SIZE = MEDIUM Status Data: { "inferenceDevice": "GPU", "inferenceLibrary": "DirectML", "canUseGPU": true, "successfulInferences": 528, "failedInferences": 0, "numInferences": 528, "averageInferenceMs": 259, "Histogram": { "car": 829, "person": 21, "bicycle": 15, "motorcycle": 2, "truck": 23 }, "NumItemsFound": 890 } Started: 18 Jul 2024 11:05:40 AM Central Standard Time LastSeen: 18 Jul 2024 11:11:15 AM Central Standard Time Status: Started Requests: 532 (includes status calls) Installation Log 2024-06-28 09:55:35: Installing CodeProject.AI Analysis Module 2024-06-28 09:55:35: ====================================================================== 2024-06-28 09:55:35: CodeProject.AI Installer 2024-06-28 09:55:35: ====================================================================== 2024-06-28 09:55:35: 310.0Gb of 487Gb available on 2024-06-28 09:55:35: General CodeProject.AI setup 2024-06-28 09:55:35: Creating Directories...done 2024-06-28 09:55:35: GPU support 2024-06-28 09:55:36: CUDA Present...No 2024-06-28 09:55:36: ROCm Present...No 2024-06-28 09:55:37: Checking for .NET 7.0...Checking SDKs...All good. .NET is 7.0.410 2024-06-28 09:55:39: Reading ObjectDetectionYOLOv5Net settings.......done 2024-06-28 09:55:39: Installing module Object Detection (YOLOv5 .NET) 1.10.2 2024-06-28 09:55:40: Downloading ObjectDetectionYOLOv5Net-DirectML-1.10.2.zip...already exists...Expanding...done. 2024-06-28 09:55:41: Copying contents of ObjectDetectionYOLOv5Net-DirectML-1.10.2.zip to bin...done 2024-06-28 09:55:43: Downloading YOLO ONNX models...already exists...Expanding...done. 2024-06-28 09:55:44: Copying contents of yolonet-models.zip to assets...done 2024-06-28 09:55:45: Downloading Custom YOLO ONNX models...already exists...Expanding...done. 2024-06-28 09:55:45: Copying contents of yolonet-custom-models.zip to custom-models...done 2024-06-28 09:55:45: Scanning modulesettings for downloadable models...No models specified 2024-06-28 09:55:48: Self test: Self-test passed 2024-06-28 09:55:48: Module setup time 00:00:10.66 2024-06-28 09:55:48: Setup complete 2024-06-28 09:55:48: Total setup time 00:00:13.09 Installer exited with code 0

Use Ctrl+Left/Right to switch messages, Ctrl+Up/Down to switch threads, Ctrl+Shift+Left/Right to switch pages.