Show HN: Cheap-IM – CPU-only voice agent approximating Thinking Machines' demo (github.com)

A new project called Cheap-IM has emerged, featuring a CPU-only voice agent that mimics the real-time capabilities of Thinking Machines' advanced Interaction Models demo from May 2026. This agent operates on standard laptop hardware and aims to achieve core interactive functions—including real-time speech processing, vision-keyed actions, live translation, and multitasking—using minimal calls to large language models (LLMs). Unlike Thinking Machines, which utilized a sophisticated 276 billion parameter model trained on continuous audio and video, Cheap-IM combines smaller, off-the-shelf models leveraging a Python event loop, showcasing impressive performance with key behaviors executed purely on a laptop. The significance of Cheap-IM lies in its accessibility and efficiency, demonstrating how common computing resources can facilitate sophisticated AI interactions without the need for elaborate infrastructure. Its technical setup includes local speech recognition (Silero VAD and Kroko ASR), object detection (YOLO11), and translation (Whisper) alongside background processing for real-time data retrieval. The orchestrator runs a single asyncio loop, allowing for streamlined event handling and interaction tracking. This project paves the way for more affordable, accessible AI solutions that can leverage existing hardware to deliver complex functionalities in voice interaction environments.