# FlashML-org/FreeToken

Repository: https://github.com/FlashML-org/FreeToken
Canonical: https://ross.abutalabs.com/products/freetoken
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2026-08-26T07:42:10+00:00

## Health v2 (maintenance only)
Score: 68/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 66, longevity 3
- inputs: {"age_days": 44, "days_push": 7, "days_rel": 14, "gap_med": null, "n_releases_24m": 1}
- flags: young
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 8370, forks 729 (observed 2026-08-28T04:10:19.803110+00:00)

## What it is
FreeToken is an edge-native Mixture-of-Experts (MoE) LLM serving engine that runs frontier-scale open-weight models on consumer hardware by treating GPUs, CPUs, and host memory as a unified inference platform. It ships as a desktop app with a GUI and exposes OpenAI- and Anthropic-compatible APIs for integration with coding agents and chat tools.

## Use cases
- run large MoE models locally on a gaming PC
- serve open-weight LLMs with an OpenAI-compatible API
- offload LLM inference between CPU and GPU
- run coding agents like Claude Code or Codex against a local model
- fit a 290B parameter model on limited VRAM
- chat with local models through a desktop GUI

## When to choose
- you want to run frontier-scale MoE models on consumer NVIDIA RTX GPUs
- you need local LLM serving with OpenAI/Anthropic-compatible endpoints
- you have limited VRAM and need elastic CPU-GPU memory offloading
- you want semantic caching to avoid recomputing context for agentic workflows

## When to avoid
- you need multi-node datacenter-scale serving rather than single-machine edge inference
- you rely on non-NVIDIA GPUs, since support focuses on RTX 30/40/50 series
- you need dense (non-MoE) model architectures that may not be supported
- you want a fully managed cloud inference service

## Facets
- artifact type: application
- maturity: active
- function: llm-inference, caching, http-server, gpu-computing
- domain: large-language-models, machine-learning, developer-tools, self-hosted
- platform: windows, python, cross-platform
- tags: mixture-of-experts, local-llm, cpu-gpu-offload, openai-compatible-api, anthropic-compatible-api, quantization, edge-inference, consumer-gpu, linux, desktop, gpu

## Member repositories
- FlashML-org/FreeToken (main) score 68

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:10:19.803110+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T17:27:33.288509+00:00, confidence not recorded.
  - readme: https://github.com/FlashML-org/FreeToken (fetched 2026-08-28T04:10:19.803110+00:00, sha d55e1879d0e3)
  - registry_pypi: https://pypi.org/pypi/freetoken/json (fetched 2026-08-29T08:27:22.939371+00:00, sha 8aa69b6f9ba1)
- Data as of 2026-08-30T08:39:29.467469+00:00.