FreeToken
Run frontier LLMs locally on your PC with optimized MoE serving
Visit flashml.ai ↗What is FreeToken?
FreeToken is a local inference engine for running large open-weight models on consumer hardware like gaming PCs and laptops. It handles Mixture-of-Experts (MoE) models up to 290B+ parameters through bandwidth-adaptive CPU-GPU co-execution, semantic-aware KV caching, and dynamic memory management. Built with claude-code, it supports models like DeepSeek-V4-Flash and Qwen, with APIs compatible with Anthropic and OpenAI for integration with coding agents and tool-calling workflows.
Designed for developers and researchers who want private, offline access to frontier models without cloud dependencies. Runs on Windows and Linux with support for various quantization formats and consumer GPUs.
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 FreeToken automatically. Evidence: Agent files (claude-code) in FlashML-org/FreeToken. View source.
Similar vibe-coded apps
#1 this weekOpen-source local voice cloning, dubbing, transcription, and audiobook creation
Clone viral videos with AI agents, generate 100 variants in one command
Desktop app for DeepSeek Harness with built-in model provider support
AI agent controls your iPhone through a $7 ESP32 board, no jailbreak needed
Local-first memory storage and retrieval system for AI workflows