PrismML Bonsai
Ultra-compressed AI models for phones, laptops, and efficient data centers
Visit prismml.com ↗What is PrismML Bonsai?
PrismML builds highly compressed language models using 1-bit and ternary quantization techniques. Bonsai 2 27B, their latest release, achieves 98.2% of full-precision performance at one-ninth the size (5.9 GB), enabling advanced reasoning, multimodal understanding, and tool use on local devices. The models support vision capabilities, agentic tool calling with OpenAI-style interfaces, and extended 262K-token context windows.
Designed for developers and organizations needing capable AI without cloud dependencies, Bonsai models run across multiple platforms including Mac, Linux, Windows, and various GPU backends. The project prioritizes intelligence density—reasoning quality per parameter—over raw model scale, grounded in efficiency research from Caltech.
Momentum
Daily activity over the last 30 days: upvotes, unique visits, new GitHub stars and Hacker News points.
How do we know it's vibe coded?
vibetrend discovered PrismML Bonsai automatically. Evidence: GitHub PrismML-Eng/Bonsai-demo; repo has AGENTS.md. View source.
Similar vibe-coded apps
Turn your codebase into a queryable knowledge graph for AI coding assistants
Operating layer for AI coding agents with skills, memory, planning and security
Self-hosted AI agent with persistent memory and multi-platform messaging
Open-source AI platform for short-form video and animated series creation
Open-source canvas workbench for AI-powered image and video creation
Community marketplace for DeepSeek Harness Web themes, pets, plugins
Open-source, AI-native knowledge base with Evernote-style interface