NevTan AI
NevTan AI is a coding agent for VS Code. Describe a task and it reads your
project, plans the change, edits files and runs commands, asking for your
approval before it touches anything.
Use NevTan's hosted coding models, run a free model on your own computer with
no API key, or combine the two.
What it can do
- Work across your whole project: it explores files and searches code to
get up to speed, then makes multi-file changes you review as diffs.
- Run commands: builds, tests and scripts run in your terminal, with their
output read back so it can fix what failed.
- Plan, then act: in Plan mode it investigates and proposes a plan
without changing anything; switch to Act when you are happy with it.
- Stay in control: every file change and command waits for your approval
unless you choose to auto-approve it. Checkpoints let you roll a task back.
- Use your selection: select lines in the editor and they are attached to
your next message automatically (shown as a chip above the input; click × to
leave them out). Cmd/Ctrl + ' adds a selection to the chat by hand.
- Extend it: MCP servers, rules and workflows work as you would expect.
Choose where the model runs
NevTan Cloud
Paste your NevTan API key (create one on the API keys page of the NevTan AI
app) and use NevTan's hosted coding models. Usage is recorded per key, so you
can see what you have used.
This computer (NevTan Local)
Choose Run on this computer in the NevTan AI settings, or run
NevTan AI: Set Up Local Model from the Command Palette. NevTan AI shows its
list of open coding models and marks which ones fit your computer's memory and
free disk space. Pick one and it is downloaded, checked and started for you.
- No API key and no account needed.
- Your code and prompts stay on your machine.
- Models are stored in
~/.nevtan/models and shared with the NevTan desktop
apps, so nothing is downloaded twice.
| Model |
Download |
Memory needed |
| Qwen3 Coder 30B |
17.3 GB |
32 GB |
| Qwen3 32B |
18.4 GB |
48 GB |
| Qwen3 30B |
17.3 GB |
32 GB |
| Muse Glimmer 30B |
16.1 GB |
32 GB |
| Gemma 4 26B |
15.8 GB |
24 GB |
| Devstral Small 24B |
13.3 GB |
24 GB |
| gpt-oss 20B |
11.3 GB |
24 GB |
| Qwen3 14B |
8.4 GB |
24 GB |
| Qwen3 8B |
4.7 GB |
16 GB |
| Qwen3 4B |
2.3 GB |
any |
| Qwen3 1.7B |
1.7 GB |
any |
All are released under the Apache 2.0 licence. Local models need macOS,
Windows or Linux on a 64-bit processor.
Hybrid mode
Run NevTan AI: Use Hybrid Mode (or choose Hybrid mode in the NevTan AI
settings). NevTan Cloud's larger model plans in Plan mode; the model on your
computer carries the plan out in Act mode. When you switch to Act, only your
request and the finished plan are handed to the local model, so it starts
quickly and follows a short, clear plan.
Other providers
You can also use your own key with Anthropic, OpenAI, Google Gemini,
OpenRouter, DeepSeek, AWS Bedrock or LiteLLM, or a local server such as
Ollama, LM Studio or any OpenAI-compatible endpoint.
Getting started
- Open NevTan AI from the activity bar.
- Choose NevTan AI and paste your API key, or choose
Run on this computer.
- Describe what you want done.
Commands
| Command |
What it does |
| NevTan AI: Set Up Local Model |
Choose, download and start a model on this computer |
| NevTan AI: Use Hybrid Mode (Cloud Plans, This Computer Acts) |
Plan with NevTan Cloud, act with your local model |
| Add to NevTan AI (Cmd/Ctrl + ' with a selection) |
Add the selected code to the chat |
| Jump to Chat Input (Cmd/Ctrl + ') |
Focus the chat box |
Privacy
With NevTan Local, requests go only to the model server on your own computer
(127.0.0.1). With NevTan Cloud, requests go to NevTan's servers; with another
provider, to that provider. The extension sends no telemetry.
Support
Questions and feedback: nevtan.com
NevTan AI is based on the open-source Cline
agent (Apache 2.0).
