The AI IDE that never phones home.
Tab autocomplete, chat and AI edits powered by open-weight models running on your computer. No cloud, no API keys, no limits. Works on a plain laptop CPU.
Free download. ~175 MB installer, then a one-time 645 MB model download. No account, no API key. Install guide · All releases
// utils.ts — typing stops, BdebTech AI suggests in ~0.5 s on CPU
export function isEven(n: number): boolean {
return n % 2 === 0;
}
export function isOdd(n: number): boolean {
return !isEven(n);
}
// local · Qwen2.5 Coder 0.5B · 53 tokens/s · 0 bytes sent to the internetWhy local
Private by architecture, not by policy
Cloud AI editors promise not to train on your code. BdebTech AI IDE cannot: the model never leaves your machine, so neither does your code.
Tab autocomplete
Fill-in-the-middle completions that see the code before and after your cursor. Single-line when you are mid-statement, multi-line at line ends.
Chat with your code
Attach the current file or a selection. Explain, find bugs, write tests. Copy, insert or apply code from the answer with one click.
Edit with AI
Select code, describe the change, review a real diff, accept or discard. Nothing is written to your file until you say so.
Runs on any PC
The Starter model answers at ~50 tokens/s on a 2021 i5 laptop with no GPU. Hardware is profiled on first run and bigger models are offered when your machine can take them.
Fully offline
One download, then zero network calls. Flights, trains, NDAs, air-gapped labs, countries where cloud AI is blocked or priced out: it just works.
Open API for other tools
An OpenAI-compatible endpoint on 127.0.0.1:41337. Cline, Continue, Aider and your own scripts can share the same local model.
How it works
From download to first completion in about three minutes
Install the IDE
A familiar VS Code-style editor (built on VSCodium) with the BdebTech AI extension and the local AI daemon pre-installed. No account.
Download the Starter model once
On first launch the IDE profiles your CPU, RAM and GPU, fetches the matching llama.cpp engine and the 645 MB Qwen2.5 Coder 0.5B model, and verifies checksums.
Start typing
Autocomplete appears as grey ghost text. Press Ctrl+Alt+L to chat, Ctrl+Alt+K to edit a selection.
Scale up when you want
Switch to the 1.5B Lite+ model for free, or unlock 7B-32B Studio packs with GPU acceleration. The model switcher shows what your hardware can run.
Comparison
Local-first versus cloud AI editors
| BdebTech AI IDE | Cursor / Windsurf | GitHub Copilot | |
|---|---|---|---|
| Code stays on your machine | Always | No, sent to cloud | No, sent to cloud |
| Works offline | Yes | No | No |
| Account or card required | No | Yes | Yes |
| Usage limits on the free tier | None | Yes | Yes |
| Model quality | Small local (0.5B-1.5B free, up to 32B on Studio) | Frontier cloud models | Frontier cloud models |
| Runs on a CPU-only laptop | Yes | N/A (cloud) | N/A (cloud) |
| Payment options | UPI, USDT/USDC, Binance Pay | Card | Card |
Honest note: a 0.5B local model is not GPT-class. It is fast, private and free. Use the Studio models on a GPU when you need more reasoning.
Pricing
Free to use. Studio for professionals.
Free
$0 forever
- Branded IDE download (Windows now, Linux and macOS next)
- Starter model: Qwen2.5 Coder 0.5B, runs on any CPU
- Lite+ model: Qwen2.5 Coder 1.5B
- Tab autocomplete, chat, Edit with AI
- Fully offline after first download
- OpenAI-compatible local API for other tools
Studio
$200 per month
- Everything in Free
- Studio model packs: Qwen2.5 Coder 7B, 14B and 32B
- GPU acceleration profiles (CUDA, Vulkan, Metal) tuned per machine
- Monthly curated model-pack updates benchmarked on real coding tasks
- Priority support from the developer on WhatsApp and email
- Up to 3 machines per license, 30-day offline grace
- Pay with UPI (India) or USDT / USDC / Binance Pay (worldwide)
FAQ
Frequently asked questions
Does BdebTech AI IDE really work offline?
Yes. After the one-time download of the engine and the Starter model (645 MB), autocomplete, chat and AI edits run entirely on your computer. You can unplug the network and keep coding. Nothing is sent to BdebTech or any cloud provider.
Will it run on my laptop without a GPU?
Yes. The default Starter model (Qwen2.5 Coder 0.5B) is chosen specifically for CPUs. On a 2021 Intel i5 laptop it answers at about 50 tokens per second with a first token in under 200 ms. 4 GB of RAM is enough; 8 GB lets you step up to the 1.5B Lite+ model.
How is this different from Cursor, Copilot or Windsurf?
Those products run frontier models on their servers, need an account and a card, and send your code to the cloud. BdebTech AI IDE runs open-weight models on your machine, needs no account, and the free tier has no usage limits. The trade-off is model size: a local 0.5B-1.5B model is faster but less capable than a cloud frontier model. The Studio plan adds 7B-32B models that narrow that gap when you have a GPU.
What is the Studio plan and why $200 per month?
Studio unlocks the larger model packs (7B, 14B, 32B), GPU-tuned engine profiles, monthly curated model updates benchmarked on real coding tasks, and direct priority support from the developer. It is priced for professionals who bill for their time and want a private, unlimited assistant; the models themselves are free and open, you pay for the curation, tooling and support. Pay with UPI in India or USDT/USDC/Binance Pay anywhere.
Can other tools use the local model?
Yes. The daemon exposes an OpenAI-compatible API at http://127.0.0.1:41337/v1. Point Cline, Continue, Aider, Open WebUI or curl at it and they share the same local model.
Is the code open source?
The editor extension and the local daemon are Apache-2.0. The editor itself is VSCodium (MIT), the engine is llama.cpp (MIT), and the models are Qwen2.5-Coder (Apache-2.0). Studio model packs and the license service are proprietary.
Which operating systems are supported?
Windows 10/11 x64 today. Linux (x64) and macOS (Apple Silicon and Intel) builds of the branded IDE are next; the extension and daemon already contain the engine builds for them.
Install it. Unplug the network. Keep coding.
Built by a working developer for developers who care where their code goes.
Free download. ~175 MB installer, then a one-time 645 MB model download. No account, no API key. Install guide · All releases