Live
ready
A2I រត់ក្នុងឧបករណ៍អ្នកផ្ទាល់ — គ្មាន API ខាងក្រៅ · runs on your own device, no external AI APIs
Load source files, then ask questions about them or ask for changes. This uses your local A2I Core — pick it in Settings first. Files stay on your machine; edits are shown here for you to download, never written for you.
Paste your own key — free from aistudio.google.com/apikey. Stored only in this browser, sent directly to Google.
Keep a GGUF model on your SSD and load it straight from disk — no download, works offline, and new preview URLs never re-download. Big models fit too. Get GGUF files from Hugging Face (Qwen, Llama, …).
| Device | Model | Size |
|---|---|---|
| 📱 Phone · 2–4 GB | Qwen2.5 0.5B–1.5B (in browser) | 1–2 GB |
| 💻 Laptop · 8 GB | Qwen2.5 7B · Llama 3.1 8B · DavidAU Fable 9B | 5–6 GB |
| 🖥️ PC · 16–32 GB | 14B–32B models | 9–20 GB |
| 🎮 GPU | same models, much faster (WebGPU / vLLM) | — |
| ☁️ No good hardware | Groq · Hugging Face (70B, free API) | — |
Rule: biggest model that fits your RAM = best you can run locally. Full list & downloads: MODELS.md
Famous, strong models for free: get a key, paste it, Add. Keys stay in this browser.
• Groq — console.groq.com/keys (Llama 3.3 70B, very fast)
• OpenRouter — openrouter.ai/keys (DeepSeek R1, Llama 70B — :free)
• Hugging Face — huggingface.co/settings/tokens (any HF model via hf_… token). Paste the newest repo id from huggingface.co/models — e.g. Llama 4, Qwen3, DeepSeek-R1
• Local, no API at all: A2I Core (any machine), vLLM (GPU, fastest), Ollama