# I built a local-first AI development environment for VS Code, Visual Studio and Eclipse
Over the last months I have been building AI Chat 2026, a desktop AI environment for developers.
The original motivation was fairly simple:
I did not want my development workflow to depend on a single AI provider, a single model, or even a single IDE.
I wanted one application where I could decide for myself whether a task should run against a local model or a cloud model — and still be able to work directly with my development projects.
That eventually turned into AI Chat 2026.
Local models and cloud APIs in one application
AI Chat 2026 can work with local AI models as well as remote providers.
The idea is not to decide for the developer which model is "the right one".
Depending on the task, I might want:
- a completely local model for private source code
- a larger cloud model for a difficult coding problem
- another provider for comparison
- an OpenAI-compatible endpoint running somewhere in my own infrastructure
The model should be replaceable.
The development workflow should not be.
That is one of the core design principles behind the application.
The IDE is not the AI backend
Another decision I made was to keep the actual AI environment outside the IDE.
AI Chat 2026 is a standalone desktop application.
The IDE integrations act as bridges between the development environment and AI Chat.
At the moment I support:
- Visual Studio Code
- Visual Studio 2026
- Eclipse
This allows the same AI configuration and model infrastructure to be used across different IDEs.
The IDE plugin provides the project and workspace context, while AI Chat handles the interaction with the selected model.
More than copying code from a chat window
One thing I wanted to avoid was the typical workflow of:
- Ask an AI for code
- Copy the result
- Find the correct file
- Paste it
- Fix whatever broke
The developer integration is designed so that AI Chat can work with the actual workspace and perform structured file changes.
The basic flow is:
AI Chat → requested change → IDE bridge → editor/workspace → file update
The integration can provide project context, read files and apply changes back into the development environment.
I am still improving this workflow, especially around larger multi-file tasks, validation and local-model tool calling.
Local AI is an important part of the project
Local inference is not an afterthought for me.
There are many situations where source code, architecture documents or internal project information simply should not leave the machine or company infrastructure.
At the same time, local models have very different hardware requirements and capabilities.
So AI Chat 2026 is intended to make it possible to combine both worlds instead of forcing one approach:
local when local makes sense, cloud when cloud makes sense.
The developer remains in control of that decision.
Why three IDEs?
Visual Studio Code was the obvious first integration.
Visual Studio 2026 followed because I work heavily with .NET.
Eclipse was interesting for a different reason: there is still a significant amount of professional Java development happening there, especially in enterprise environments.
Supporting all three also forced me to separate the actual AI workflow from IDE-specific implementation details.
That was useful architecturally.
The Eclipse integration, for example, uses a local bridge and automatically reconnects when AI Chat is started after Eclipse.
The goal is that the IDE connector becomes infrastructure, not something the developer has to constantly manage.
I now need developers who will actually try to break it
At this stage I can test a lot myself, but that has an obvious limitation:
I know how the software is supposed to work.
External developers do not.
And that is exactly the feedback I need now.
So I am opening a small developer test program.
I am looking for 10 developers who are willing to actually use AI Chat 2026 and give me honest feedback.
For this round I am providing:
- 5 AI Chat 2026 Pro licenses
- 5 AI Chat 2026 ProPlus licenses
I am not looking for positive reviews.
I am interested in things like:
- What is confusing?
- What workflow feels unnecessary?
- What breaks?
- What is missing?
- Does the IDE integration behave as expected?
- How well does the workflow work with local models?
- What would prevent you from using it for real development?
There is also no requirement to publish a review or promote the product.
If a selected tester genuinely uses the software and provides useful feedback, the provided license can be kept permanently.
Application period
This test round is open from November 12 until November 22, 2026.
You can find the test program here:
More information about AI Chat 2026:
I am especially interested in feedback from developers using local models, .NET, Java, Visual Studio Code, Visual Studio or Eclipse.
And if you do not want to participate in the test program but have questions about the architecture or the approach, I am happy to discuss those as well.
Top comments (0)