🦅 PegasusLink mini

A fully offline AI for your own devices — laptop, desktop, tablet, phone. No server, no account. Runs on-device.

Version

…

🔒 By author's approval

Downloads are by the author's consent. Send a request below — once the author approves it, a personal download link (time-limited) is e-mailed to you. There are no public download links on this page.

Prefer email? Write to info@reverseml.online.

🤗 Or download directly — no approval needed

open · Apache-2.0

The model weights (pegasus-mini-q4.gguf) are public on Hugging Face. Use them straight away with llama.cpp, Ollama, LM Studio or a phone shell — no request, no wait.

Download on Hugging Face → ⬇️ Direct .gguf (986 MB)

The form below remains for the native Windows setup.exe app and personalised support.

🔒 Privacy. Trained on general knowledge only (science, code, math) — no personal data. Runs fully on your device and never sends your data anywhere.

🧰 Windows app extras. Attach files (text, code, CSV, PDF), images (on-device OCR) and audio/video (on-device Whisper transcription), then ask about them — plus an exact chemistry equation balancer that computes on-device, never guesses. All processing runs fully offline on your machine.

💻 Laptop / desktop

Windows: the approval link gives a native setup.exe app → install Ollama → see model setup below.
macOS / Linux: the link gives the model file — use it with Ollama or LM Studio.

# put pegasus-mini-q4.gguf in a folder, then:
printf 'FROM ./pegasus-mini-q4.gguf\nPARAMETER temperature 0.3\nPARAMETER num_ctx 4096\n' > Modelfile
ollama create pegasus-mini -f Modelfile
ollama run pegasus-mini "Explain a Kalman filter in one sentence."

📱 Phone / tablet (Android / iOS)

Install PocketPal AI (free, Android & iOS).
Models → Add from file → pick the downloaded pegasus-mini-q4.gguf → chat offline.

Loads on demand and unloads when idle — it won't hog RAM.

♻️ It keeps getting smarter

New versions are trained alongside the main PegasusLink and published here; your app can check /models/manifest.json for updates.

🤖 Embedded & autonomous systems

Because it runs fully on-device and offline, PegasusLink mini can serve as the on-board cognitive layer for autonomous and unmanned systems — drones, rovers, robots — turning natural-language missions into high-level plans and commands, and learning routines on the device. It is advisory only: in the reference two-tier “device brain” design it pairs with a deterministic real-time controller and a safety monitor that always retains command — the model never actuates directly.

⚠️ Experimental framework — not a certified autopilot or ADAS. You are responsible for safe, lawful operation; keep a human in command with an independent manual override.