Local AI on Fedora - Running Ollama and Obsidian Offline

How many times is it frustrating to manually copy raw chat logs from browser interfaces, paste them into text files, and clutter a vault with messy, unformatted text? Setting up a seamless bridge to capture AI conversations from a browser and turn them into clean, structured notes inside Obsidian—while keeping everything completely private—makes running a local model worth the effort.

This article explores how to turn a Fedora Linux machine into a local processing hub using Ollama, install Obsidian, integrate them directly, and bridge browser chats into notes.

However, before diving into local Large Language Models (LLMs), understanding the hardware realities is essential. Running heavy local models on older hardware, entry-level ultrabooks without dedicated graphics, or resource-constrained systems leads to frustratingly slow performance. For those use cases, leveraging a cloud API alternative (such as Google AI Studio, covered in a follow-up post) is often a much smoother choice.

Hardware Reality Check: Is Local AI Right for You?

Local model inference is fundamentally memory- and compute-bound. Before downloading gigabytes of model files, keep these guidelines in mind:

• The Graphic / Processor Bottleneck: Running models locally smoothly requires a modern multi-core processor and a decent amount of RAM or GPU VRAM. If a machine struggles with heavy multitasking or lacks proper graphics acceleration, local LLMs will crawl at a few tokens per second.

• The Disk Space Trap: AI models are multi-gigabyte files, and they accumulate quickly as different variants are tested.

  • A lightweight model (like 3B parameters) takes around 2 GB of disk space.

  • A standard mid-tier model (like Qwen 2.5 7B) takes roughly 4.7 GB.

  • Larger models (14B or 32B) easily range from 10 GB to 20 GB+.

  • Tip: I always provision at least 15 to 20 GB of free NVMe space just to safely manage the initial model library and temporary download layers.

If the system checks out and disk headroom is sufficient, it’s time to get everything running.

Step 1: Installing and Starting Ollama on Fedora

Ollama installs cleanly on Fedora using the native package manager (dnf), ensuring smooth system updates without risking downloaded model files (stored safely in user directories).

• Installation

Open terminal and install Ollama via DNF:

sudo dnf install ollama

• Enable and Start the Service

Ensure the daemon runs and stays active in the background:

sudo systemctl enable --now ollama

• Pull an Efficient Model

To find and explore available models, consult the official Ollama Library (ollama.com/library), where open-weight models (such as Llama, Qwen, Mistral, and Gemma) are cataloged with their respective sizes and parameter variants.

In a separate terminal, download a well-balanced model like qwen2.5:7b (~4.7 GB):

ollama run qwen2.5:7b

Once downloaded, a local REST API runs at http://localhost:11434.

Step 2: Installing Obsidian on Fedora

To manage notes locally using plain Markdown files, Obsidian provides the ideal environment. The cleanest approach on Fedora is via Flatpak:

flatpak install flathub md.obsidian.Obsidian

This keeps the application sandboxed while allowing full read/write access to the chosen vault directory on the local filesystem.

Step 3: Connecting Ollama to Obsidian & Bringing Browser Chats In

To solve the friction of AI-assisted note-taking practically, bridge the workflow using community tools:

1. Inside Obsidian: Open the vault, navigate to Community Plugins, and install a plugin like Text Generator to interact directly with the local Ollama instance (http://localhost:11434) without cluttering the workspace.
2. From the Browser to Obsidian: When using web-based interfaces to chat, lock-in to a single browser is avoided. Thanks to standard web extension protocols, tools work seamlessly across Firefox, LibreWolf, or Chromium-based browsers, exporting or formatting clean Markdown directly into the Obsidian vault without manual copy-pasting.

Conclusion & What’s Next

Running Ollama locally delivers a robust, private assistant on Fedora. However, if hardware isn’t suited for local inference, don’t worry: in the next article, the cloud alternative will be explored, showing how to hook up a free Google AI Studio API key to Obsidian for instant, zero-footprint workflows.