Erebrus AI

Erebrus AI runs real language models on your own device. Download an open model once, then chat, summarize and transcribe with no cloud round-trip, no account and no third party in the middle.

PRIVATE BY DEFAULT Inference happens on your hardware. Your prompts, transcripts and audio stay in the app's own storage — nothing is uploaded by any of these workflows.

CHAT WITH LOCAL MODELS

• Open models including Llama, Qwen, Gemma, Phi and Mistral, plus reasoning distills
• Live token streaming with a tokens-per-second readout
• Copy, share, regenerate or read a response aloud
• Works fully offline, in airplane mode, on a plane or in the field

A MODEL LIBRARY BUILT FOR YOUR DEVICE

• Curated catalog from 138 MB nano models up to desktop-class weights
• Every package is checksum-verified before it is allowed to run
• Erebrus picks the format and quantization your device can handle — GGUF through llama.cpp — and you can choose a different package yourself
• Resumable downloads, storage usage at a glance, updates with rollback

ON-DEVICE TRANSCRIPTION

• Record and transcribe speech locally in twelve languages using a verified local Whisper model
• Audio is saved with each session so you can replay it against the raw transcript
• Tap a sentence to jump to that moment; edit the transcript and compare against the original
• No chat model is required to transcribe, and no analysis runs automatically — you send a draft to your model only when you choose to

PERSONAS YOU CONTROL Set the system prompt, temperature, top-p, token budget and repeat penalty. Save personas privately or share them with your team.

FOR TEAMS AND ORGANIZATIONS Erebrus AI discovers other Erebrus nodes on your local network and can use models a workspace shares with you. Sign-in is optional and only needed for team workspaces — everything local keeps working without an account. The data path stays inside your own network, which is the point: sovereign AI your organization actually controls.

Part of the Erebrus ecosystem by NetSepio.