⚡ AMD  ·  Apple Silicon  ·  Intel Arc  ·  NVIDIA

Local AI +
Graphnosis job memory.
Any GPU. Zero Cloud.

Pair Graphnosis with Ollama to unlock the full Foresight tier — semantic memory enrichment, insights, distillation, foresight predictions, neural inference, and autonomous skill & goals training — on whatever hardware you already own.

Local LLM via Ollama

What local AI unlocks

Graphnosis works fully offline without a GPU. Add Ollama on any supported GPU and the Foresight tier activates — all computed locally, nothing leaves the device.

Semantic Recall Enrichment

Your recall queries are rewritten and expanded by the local LLM at search time — surfacing memories your original phrasing would have missed.

Insights & Patterns

Background analysis of your memory graph surfaces patterns, gaps, and opportunities — the hippocampus working while you focus on other things.

Knowledge Distillation

Feed raw text — meeting notes, documents, web pages — and the local LLM extracts discrete facts and routes them into your knowledge graph automatically.

Foresight Predictions

Before you act, Graphnosis surfaces risks and opportunities grounded in your actual memory — not generic LLM advice, but predictions anchored to your history.

Neural Edge Prediction

A local Graphnosis Neural Network runs in the background proposing connections between memory nodes — relationships your graph doesn't explicitly record yet.

Skill Auto-Training Pro

The local LLM rewrites and refines your Skills & SOPs in place — improving structure and clarity while preserving source attribution. All on-device.

Setup Guide

Pick your hardware

Same Graphnosis, same Ollama, same Foresight features — the only difference is which driver stack you need.

AMD · ROCm

AMD GPU

RX 6000 series  ·  RX 7000 series  ·  Radeon Pro  ·  Linux

  1. 1 Download Graphnosis for Linux (.AppImage or .deb)
  2. 2 Install ROCm drivers for your card. AMD's official guide: ROCm Quick Start
  3. 3 Install Ollama — it picks up ROCm automatically, then pull a model: ollama pull llama3.2
  4. 4 In Graphnosis → Settings → Foresight → enable Local LLM → select your model

Notes

ROCm support is Linux-only. Windows ROCm is experimental and not recommended for production use.

Supported cards: RX 6600 and later, RX 7600 and later, Radeon Pro W6000/W7000 series. Older GCN cards are unsupported by Ollama.

If ollama run llama3.2 falls back to CPU, verify ROCm is recognised with rocminfo | grep -i agent.

Apple · Metal

Apple Silicon

M1  ·  M2  ·  M3  ·  M4  ·  All variants (Pro / Max / Ultra)  ·  macOS 13+

  1. 1 Download Graphnosis for macOS (Apple Silicon native)
  2. 2 Install Ollama for macOS — no extra drivers needed. It uses Metal automatically via llama.cpp.
  3. 3 Pull a model sized for your RAM: # M1/M2 8 GB → 3B; M2 Pro+ or 16 GB+ → 8B ollama pull llama3.2 # 3B, fits any M-chip Mac ollama pull llama3.1:8b # 8B, recommended for 16 GB+
  4. 4 In Graphnosis → Settings → Foresight → enable Local LLM → select your model

Notes

Ollama uses the unified memory architecture — the GPU and CPU share the same memory pool, so 8 GB Macs can comfortably run a 3B model. 16 GB opens up 8B; 32 GB+ handles 13B models well.

No ROCm, no CUDA, no driver install required. Ollama ships its own Metal backend.

Intel Mac? CPU-only inference still works — just slower. An Apple Silicon Mac or an external GPU is recommended for the full Foresight experience.

Intel · oneAPI

Intel Arc

Arc A-series  ·  Arc B-series  ·  Windows & Linux  ·  Support growing

  1. 1 Download Graphnosis for Windows or Linux
  2. 2 Update your Arc drivers to the latest version from Intel's download centre
  3. 3 Install Ollama and pull a compact model (Arc VRAM is limited): ollama pull llama3.2:3b
  4. 4 In Graphnosis → Settings → Foresight → enable Local LLM → select your model

Notes

Intel Arc support in Ollama is actively maturing. Acceleration uses Intel's oneAPI / SYCL backend — expect occasional model compatibility gaps with very large quantisations.

Recommended: Arc A770 (16 GB) or Arc B580 for best results. Smaller cards (A380, A310) will fall back to CPU for larger models.

Linux performance is generally ahead of Windows for Arc — if you have a choice, prefer Linux.

NVIDIA · CUDA

NVIDIA GPU

RTX  ·  GTX  ·  Quadro / RTX Pro  ·  Any CUDA-capable card  ·  Windows & Linux

  1. 1 Download Graphnosis for Windows or Linux
  2. 2 Install Ollama — it auto-detects your NVIDIA GPU via CUDA. No extra configuration needed.
  3. 3 Pull a model matched to your VRAM: # 8 GB VRAM → 8B; 16 GB+ → 13B or 32B ollama pull llama3.2 # 3B, any card ollama pull llama3.1:8b # 8B, RTX 3070+
  4. 4 In Graphnosis → Settings → Foresight → enable Local LLM → select your model

Notes

CUDA is the most mature and widely tested Ollama backend. Any Kepler-architecture card or newer works; Ampere (RTX 30xx) and Ada Lovelace (RTX 40xx) deliver the best performance per watt.

VRAM is the limiting factor — the model must fit. 4-bit quantised 8B models need ~5 GB; 13B needs ~8–9 GB. Ollama offloads excess layers to CPU RAM automatically, at a speed penalty.

NVIDIA · Jetson · Edge

NVIDIA Jetson Orin

Nano Super  ·  NX  ·  AGX  ·  JetPack 6  ·  Ubuntu 22.04

  1. 1 Flash JetPack 6.x from NVIDIA — ships with Ubuntu 22.04, CUDA, and cuDNN pre-installed
  2. 2 Install Graphnosis — download the Ubuntu .deb and install with sudo apt install ./Graphnosis_*_amd64.deb
  3. 3 Install Ollama — detects Jetson GPU automatically: curl -fsSL https://ollama.com/install.sh | sh
  4. 4 Pull a compact model (Nano Super has 8 GB unified; AGX has 64 GB): ollama pull llama3.2:3b # Nano Super ollama pull llama3.1:8b # Orin NX / AGX
  5. 5 In Graphnosis → Settings → Foresight → enable Local LLM → select your model

Notes

Jetson Orin uses unified memory — GPU and CPU share the same pool. The Nano Super's 8 GB comfortably runs a 3B model; the AGX Orin 64 GB handles 13B+ models.

Ideal for air-gapped, edge, or robotics deployments where cloud connectivity is unavailable or prohibited. Graphnosis + Ollama on Jetson delivers a complete, network-isolated AI assistant stack.

Who it's for

Use cases by vertical

Robotics & Autonomous Systems

A Jetson-powered robot maintains persistent memory of its environment, tasks, and operator instructions across power cycles — no cloud, no latency, no data leak. Graphnosis is the memory layer; Ollama on Jetson is the reasoning engine.

Healthcare Edge AI

Clinical AI tools running on-device in hospitals or ambulances remember patient context, workflow history, and care protocols session-to-session — with all data encrypted locally. Private by architecture, not by policy.

Defense & Public Sector

Field AI systems that cannot touch the internet. Graphnosis + Ollama on any air-gapped GPU delivers a complete, network-isolated AI assistant stack — persistent memory, local reasoning, zero dependency on any external service.

Industrial & Manufacturing

Factory-floor AI assistants on edge hardware remember machine history, maintenance logs, and operator SOPs — with insights that surface anomalies before they become failures. No cloud connectivity required on the plant floor.

Agriculture & Remote Operations

Autonomous farm equipment, drones, and remote monitoring systems operate where connectivity is absent or unreliable. Graphnosis provides persistent local memory for AI agents that must work independently of any network.

Privacy-Conscious Professionals

Lawyers, doctors, and financial advisors get the full Graphnosis Foresight tier — semantic search, insights, distillation, and autonomous skill & goals training — on their own workstation GPU, with a compliance guarantee: data never leaves their machine.

Ready to build?

Download Graphnosis, set up Ollama on your GPU of choice, and your AI gets a private memory it keeps forever — on your hardware, under your control.

Download Graphnosis Get Ollama

Graphnosis is not affiliated with or endorsed by NVIDIA Corporation, AMD Inc., Apple Inc., or Intel Corporation. All trademarks are the property of their respective owners.