Extension for VS Code · free
AI agent for VS Code with your choice of AI model
Describe the task in ordinary language. dimosi reads the project, creates and edits files, and runs commands. It shows every change before applying it.
Choose your own AI model: Claude, GPT, DeepSeek, hundreds of models through aggregators or a free model running on your computer. You pay the provider directly using your own key. dimosi has no subscription fee.
VS Code 1.140 and latermacOS and WindowsMIT license
Three steps to the first task
- 01
Download the file
The button above downloads dimosi.vsix. This is the extension installation file.
- 02
Install in VS Code
Open the extensions panel (Cmd+Shift+X, on Windows Ctrl+Shift+X). At the top, click "...", then Install from VSIX and select the downloaded file. The dimosi icon will appear on the left.
- 03
Connect an AI model
Open your project folder and click on the dimosi icon. Select the service, paste the API key and select a model from the list.
You only need to install it manually once. Then dimosi updates itself and installs only versions with my digital signature.
What you control
The agent works with your files and spends your money. That makes the controls you keep in your hands the most important part of dimosi.
Every edit
Changes appear in the chat as a colour-coded before-and-after comparison with “Применить” (“Apply”) and “Отклонить” (“Reject”) buttons. After a task, you can see the changed files and revert any one of them or all changes together.
Plan before work
In “Сначала план” (“Plan first”) mode, the agent examines the task and presents a plan without changing anything. It starts work only after you click “Выполнить план” (“Run plan”).
Task budget
You can see the cost as a task runs and set a spending limit for it. When that limit is reached, dimosi stops and shows the amount spent. It also stops the model if it repeats the same action three times in succession.
Keys and secrets
Keys are stored in the operating system’s secure storage: Keychain on Mac and Credential Manager on Windows. The agent does not read password files such as .env. On Mac, its commands run in a sandbox without access to your personal folders.
Risky commands
Deleting folders, pushing code to GitHub and running sudo commands require separate approval every time, even if you have allowed other actions without confirmation.
Project Rules
Write down your style, prohibitions and terms in a file. The agent rereads the rules before each response. General rules apply to all projects; folder rules apply only to this folder.
What else it can do
- Understands screenshots. Paste a screenshot from the clipboard and the agent will see the error.
- Reads documentation on the Internet, but only on sites that you allow.
- Creates images for the project. This requires a Polza AI key. Each image has a separate charge and requires approval.
- Delegates initial research to a second, cheaper model. That assistant has read-only access.
- Detects errors reported by the editor after its changes and fixes them.
- Remembers the last 30 chats in each folder, you can return to any of them.
Services
dimosi is not tied to a single provider. You can change the provider or model at any time and store keys for several providers at once.
- Anthropic
- Claude models directly
- OpenAI
- using an API key or your ChatGPT subscription
- DeepSeek
- DeepSeek models
- OpenRouter
- international model aggregator
- TeamoRouter
- Claude, GPT and Gemini
- Polza AI
- model aggregator, payment in rubles
- Ollama
- free, with the AI model running on your computer
- Your own server
- any service compatible with OpenAI API
I will configure dimosi for your team
You can download and install it yourself for free. If you want clinic or agency staff to use the agent, I can help set it up for the team.
I choose a provider and model to fit your budget, write rules for your projects and show the team how to assign tasks. We start with a short conversation about what you want the agent to handle.
Frequently asked questions
Do you need to know how to program?
No. Describe the task in ordinary language, such as “make a page with a contact form”. The dimosi guide is written for non-programmers and explains each action step by step.
How much does it cost?
The extension itself is free. You only pay your chosen AI provider, using your own key. dimosi shows the cost of each task. With Ollama, there is no provider charge because the model runs on your computer.
Where can I get an API key?
In your chosen provider’s account, under API keys. The key is kept in the operating system’s secure storage and can be transferred to another computer in an encrypted file.
Can the agent break something?
You see every change before it is applied and can reject it. You can revert a task’s changes file by file or all at once. Deletion, pushing code and sudo commands require separate approval. Start with a separate test folder.
Is it possible to give the agent patient data?
You should not. File content read by the agent is sent to your chosen AI provider. Keep patient data outside the project folder. The exception is Ollama: it runs on your computer, so data stays there.
How do updates work?
dimosi checks for a new version every six hours and installs it only if the update is signed with my key and the file checksum matches. Keys, rules and settings are saved during the update. Auto-update can be turned off in the settings.
Does it work on Windows?
Yes. The only difference is that command sandboxing is available only on Mac. On Windows, each command requires approval instead.
Is it possible to work without VS Code?
Yes. dimosi has a terminal version with the same rules and approval controls. Install it through npm and run the dimosi command.
Is something not working?
In VS Code, press Cmd+Shift+P and select “dimosi: Сообщить о проблеме” (“Report a problem”). Copy the report and send it to me on Telegram @dim_osi. The report contains no API keys or conversation text.
Terminal version: dimosi-cli.tgz, install with a command npm install -g ./dimosi-cli.tgz.
Implementation
Set up dimosi for your team
Tell me who will use the agent and what tasks you want it to handle. I reply within one business day.