Kodnik works directly on your project files, runs commands and carries a task from your instruction to a finished change. It runs on Windows — locally, on your computer.
Not sure how Kodnik will run on your computer? Download the DEMO and test local models and project work before purchasing.
Kodnik runs local AI models and gives them tools to work with your project. From the app, you can create new files, edit existing ones, analyze code and fix errors.
Start with a short instruction, and Kodnik takes the task from an idea to a working change in your files.
Kodnik recognizes whether the task is to create a program, edit an existing project or fix an error. It shows a work plan, applies file changes and lets you return to the previous state with one click.
From a short instruction, through a work plan, to a concrete change in your files. You see every step and can immediately request further changes.
Work on a project folder, control changes, undo edits and receive updates — all locally, on your computer.
Before Kodnik edits or fixes a file, it saves a restore point. After the work is complete, you see a change notice below the code and can use Undo change to restore the file to its state before the edit.
You and the AI agent use the same project terminal session. You can enter commands manually, and Kodnik can continue working in that same PowerShell, CMD or Git Bash session — running commands, reading results and reacting to errors.
The most important local-model settings are available from Kodnik's bottom bar.
Small context keeps work lighter, large context gives the model more room for files and conversation, and Auto manages the setting automatically. The bottle shows how much of the selected context is currently in use — it fills as usage grows, while the percentage below shows the current level.
Use Thinking for a more demanding task, or Fast mode when you want a shorter model response.
On first launch, Kodnik detects available resources, selects the compute backend and distributes model work between the GPU, system RAM and CPU. You do not need to configure runtime parameters manually.
Detects a supported GPU, available VRAM, driver version, CPU core count and system RAM before starting the model.
Automatically selects the appropriate GPU acceleration backend — CUDA or Vulkan — or CPU mode when a supported GPU is not available.
Calculates how many model layers can be placed in GPU memory and how many should remain in RAM, using available resources without overfilling memory.
Kodnik can work with GPUs from different manufacturers and can also run on CPU only when needed.
Reads available system memory and GPU memory to configure runtime settings without manual parameter entry.
If every hardware parameter cannot be detected, Kodnik uses conservative startup settings instead of relying on incomplete data.
Kodnik provides a library of local AI models. You choose models directly in the app from the AI Models panel. The library will be expanded with future Kodnik updates.
Q6_K_L · Q5_K_M · Q4_K_M
Q3_K_L
Q2_K
In the AI models panel, you can use AUTO, ONE MODEL and TEAM.
Kodnik is a Windows desktop application. In OFFLINE mode, model and project-file work stays local. ONLINE mode enables internet tools; a connection is also needed to download application updates.
Work with project files without a cloud account and without a permanent internet connection.
After switching to ONLINE, Kodnik can search the web, open and read pages and documentation, and download files needed while working on the project.
Kodnik does not require a cloud account. In OFFLINE mode, the model runs on your computer and the project is not sent to an external AI model. You deliberately enable ONLINE mode when you want to use internet tools.
Open a project folder, enter an instruction and see changes in files on your computer — from the idea and file analysis to concrete project changes.
Memory requirements depend on the selected local model. Kodnik can run on CPU only or use a compatible GPU and automatically choose how many model layers to offload to VRAM.
| Component | Requirement | Information |
|---|---|---|
| System | Windows 10 / 11 64-bit | Kodnik is a desktop application installed on a Windows computer. |
| CPU | x86-64 with AVX2 support | Required to run local models in the current version. |
| RAM | Depends on the selected model | Different models use different amounts of memory. If you are not sure which option is suitable for your computer, you can test Kodnik with the DEMO. |
| Graphics card | Optional | Kodnik can run without a dedicated graphics card, but a compatible GPU can significantly speed up local model responses. |
| VRAM | Kodnik uses available GPU memory | Kodnik can place part of a model in VRAM and run computations on the GPU. This reduces CPU and system-RAM load and speeds up model responses. |
| Disk | Depends on selected models | Kodnik downloads the selected model files locally. Required free space depends on the number and size of the models you choose to download. |
| Internet | Needed for setup and updates | Required for initial setup, model downloads, updates and ONLINE mode. |
Kodnik configures CPU, RAM and VRAM use for the running model. Not sure how it will perform on your computer? Download the DEMO and test it before purchasing.
The DEMO lets you run Kodnik on your own hardware, download a selected local model and test real-world performance, normal chat, project and code work, the terminal and ONLINE mode.
Time is counted only while Kodnik DEMO is running. When the app is closed, the timer stops and continues on the next launch. The Thinking feature is available in the full version of Kodnik.
You receive the full Windows application and frequent updates at no extra cost. From the AI Models panel, you choose which models you want to download and use locally.
One-time payment: USD $5.99
Go to secure checkoutCheckout is handled securely by Paddle. The price is USD $5.99 as a one-time payment — there is no subscription.
After a successful payment, Kodnik automatically sends a private download link to the email used at checkout. The link is valid for 72 hours and allows up to 5 download attempts.
Already purchased? Get a fresh private download link. Questions about Kodnik: kontakt@kodnik.pl.
Open a project folder and enter an instruction. Kodnik reads your files, shows a plan and then applies code changes — it can create new files, edit existing ones and fix errors. It saves a restore point before each change.
Internet is needed for initial setup and model downloads, and later for updates and ONLINE mode. Working with a local model in OFFLINE mode does not require a permanent connection.
In OFFLINE mode, the model runs locally on your computer and the project is not sent to an external AI model. You deliberately enable ONLINE mode when you want to give Kodnik access to internet tools.
Yes. A graphics card is not required. Kodnik can run on the CPU, while a compatible GPU can significantly speed up local model responses.
No. The purchase covers the Kodnik application. The models are independent projects distributed under their own licenses. Kodnik lets you choose supported models, downloads their files and prepares them for local use.
No. You pay once — with no subscription and no cloud account. Frequent updates are included at no extra cost.
Yes. Kodnik has a built-in updater and notifies you when a newer version is available. Updates are included at no extra cost.
Installing the application itself takes a few minutes. AI models are downloaded locally based on your selection, so download time and required disk space depend on the selected models and your connection speed.
Download the 2-hour DEMO. On your own computer, you can select a model, run it locally and test actual performance before purchasing the full version.
Context controls how much information is available to the model during a task. You can choose small, large or Auto. Thinking mode is intended for more demanding tasks, while Fast mode produces shorter model responses.
