Open-source AI agent optimized for small local models (9B). Chat via terminal, browser, or Telegram. Semantic memory, 32+ tools, RAG, skill creation from chat. Runs on gaming laptops (8GB GPU). No cloud, no API keys, no subscriptions. MIT license.
Hey Product Hunt! My name is Kir, I'm the maker of Qwe-Qwe.
I built this because I was spending $300/month on cloud AI subscriptions while my laptop RTX 4070 8Gb sat idle during non-gaming hours. Modern 9B models like Qwen 3.5 are surprisingly capable — but a raw model is just a chatbot. I wanted a real assistant.
The core philosophy: don't try to make a 9B model as smart as GPT or Claude. Instead, build smart infrastructure around it. When the model generates broken JSON, repair it. When a tool fails, retry with a different approach. When the context fills up, summarize and save to memory. When a task succeeds, remember how for next time.
The result: an agent that runs 100% on my laptop, remembers my preferences across sessions, manages files, runs shell commands, indexes my documents, and creates its own skills when needed.
What's next: voice interface, and MCP protocol support.
Would love to hear your feedback! Join us at @qwe_qwe_ai on Telegram.
Co-Founder AI