A designer recently replaced their $20-per-month Claude Pro subscription with a free, locally-run setup that combines three open-source tools: LM Studio for running AI models on their computer, Fabric for applying pre-written system prompts, and Obsidian for organizing outputs. The setup works entirely offline and costs nothing, offering a practical alternative for anyone seeking privacy, reliability, or simply wanting to avoid recurring subscription fees.

What Makes This Setup Work as a Claude Projects Replacement?

Claude Projects, Anthropic's paid feature, lets users create containers with preloaded context that persist across multiple conversations. The appeal is clear: you can load research, design briefs, or project notes once, then reference them repeatedly without recopying information. The challenge with local AI is that most tools lack this kind of workspace structure. That's where Fabric enters the picture.

Fabric is a command-line tool created by Daniel Miessler in early 2024 and rewritten from Python into Go for better performance. It works by applying pre-written system prompts, called Patterns, to any text or web content you feed it. The tool comes with over 250 Patterns built in, ranging from summarization and key-point extraction to more specialized tasks like rating content quality, writing in a particular voice, or pulling claims from a piece of writing. Each Pattern is a readable markdown file that users can customize.

The designer's workflow looks like this: LM Studio runs a local language model in the background, Fabric sends prompts to that model via the command line, and Obsidian captures the outputs as searchable markdown notes. The structure mirrors Claude Projects because Obsidian acts as the persistent knowledge layer, while Fabric handles the instruction-following. One key difference is that Fabric is "one-shot," meaning it doesn't maintain chat history. But that limitation becomes a feature when combined with Obsidian: every output lands as a regular note that can be tagged, linked, and organized alongside research and inputs.

How to Build Your Own Local AI Workspace?

• Install LM Studio: Download and run LM Studio, which serves as the local AI engine. Load a model like Gemma 4 E4B or Qwen 3.5 9B, both of which handle document analysis and synthesis well on consumer hardware.
• Set Up Fabric: On Windows, install Fabric with a single command (winget install danielmiessler.Fabric). Run fabric --setup to configure your AI provider and download the Patterns library. The interactive setup walks you through each step without requiring terminal expertise.
• Connect to Obsidian: Point Fabric's output flag (-o) to a folder inside your Obsidian vault. Obsidian automatically picks up new markdown files, making every Fabric output instantly searchable and linkable within your notes.
• Organize with Wikilinks: Structure your vault with separate folders for inputs and outputs, connected by wikilinks. This creates the persistent context layer that Fabric itself doesn't provide, mimicking the workspace coherence of Claude Projects.

The designer noted that "Patterns handle the instructions, Obsidian handles the knowledge, and my local model handles the responses, and of course, it's all local and usable offline." This three-tool approach costs nothing and requires no subscription.

Why Local AI Became Essential During an Internet Outage?

The real-world value of this setup became apparent when the designer experienced a two-day internet outage combined with cell tower failures. During that stretch, cloud-based tools like Claude and NotebookLM were unreachable, but the local setup kept working. The designer was able to continue analyzing research PDFs, working on design projects, and taking notes without interruption.

This isn't unique to one region. Internet reliability is often taken for granted, but knowledge workers regularly encounter scenarios where connectivity fails: hotel WiFi throttling during peak hours, Airbnb routers requiring reboots with unresponsive hosts, or scheduled power outages that affect both internet and cell service. Most people lose productivity during these gaps. With a local LLM setup, work continues.

The designer tested two models during the outage: Gemma 4 E4B and Qwen 3.5 9B. Qwen 3.5 9B excelled at document synthesis, while Gemma 4 E4B handled visual analysis and image-based tasks better. Both ran smoothly on the designer's hardware without requiring cloud connectivity. For even more limited scenarios, like when power outages also knocked out desktop access, the designer had PocketPal with Gemma 4 E2B running on a phone, enabling quick lookups and research synthesis from mobile.

What Are the Practical Limitations of This Approach?

The local setup doesn't replicate every aspect of Claude Projects. The most obvious gap is that local models lack the reasoning power of frontier models like Claude 3.5 Sonnet or GPT-4. The designer acknowledged that "the model isn't as strong as Claude, obviously, but the structure is there." For tasks requiring cutting-edge reasoning, a local 9-billion-parameter model will fall short. However, for summarization, extraction, synthesis, and structured analysis, local models perform adequately.

Another consideration is that Fabric's one-shot nature means no persistent chat history. Users must manually structure their vault to maintain context across sessions. This requires more deliberate organization than Claude Projects, where context lives inside the project container automatically. However, this friction also encourages better note-taking habits and creates a searchable archive of all AI-assisted work.

The setup also depends on having adequate local hardware. A modern laptop with a decent GPU or neural processing unit can run a 9-billion-parameter model, but older machines may struggle. The designer's choice of Gemma 4 E4B and Qwen 3.5 9B reflects models specifically optimized for consumer hardware, but performance will vary based on the device.

When Should You Consider a Local AI Workspace?

Local AI isn't necessary for most users. Cloud tools like Claude, ChatGPT, and NotebookLM offer superior reasoning, faster responses, and no setup friction. However, local AI becomes valuable in specific scenarios: unreliable internet connectivity, privacy concerns, offline work requirements, or simply the desire to avoid subscription costs. The designer's experience shows that local AI shines during outages, but it also works as a daily tool for privacy-conscious workflows or for users in regions with unstable connectivity.

The broader trend is that local LLMs have become increasingly accessible over the past six months. Models are now optimized for phones and consumer laptops, runners are improving, and tools like Fabric and Obsidian make integration straightforward. The designer's setup proves that a free, offline alternative to paid cloud AI is no longer a technical novelty; it's a practical option for specific use cases.