01
Model
More model. Less memory.
Quantization reduces the memory needed for model weights. Explore our FP8 and NVFP4 releases.
Explore open models
We develop local AI applications and services: open models quantized for everyday hardware, self-hosted AI agent tools, and mobile apps that connect to your own AI.
Explore our products01
Model
Quantization reduces the memory needed for model weights. Explore our FP8 and NVFP4 releases.
Explore open models02
Agent
Run Hermes on your server. Choose the model provider it uses for conversations and file analysis.
See how Muninn connects03
App
Muninn connects your iPhone to the Hermes agent on your own server.
Meet Muninn

No. 01
Your iPhone, connected to the Hermes AI agent you run at home. A paired P2P connection puts your own server within reach.
Meet MuninnNo. 02
Quantized open models
We quantize open models into formats such as FP8 and NVFP4 so they can run on a single machine or GPU. Our models on Hugging Face have been downloaded more than 780,000 times.
As of November 2026.
Explore the models
Illustrative weight memory comparison
No. 03
An iPhone app for tracking options credit spreads. See your open and closed positions and profit and loss at a glance.
Meet BPSTracker
With Muninn, your messages go to your paired Hermes server. Hermes then uses the AI provider you configure, which may be a cloud service. We do not receive or store your conversation content. Read the Muninn privacy policy