How To Install Fabric - Open-Source AI Framework That Can Automate Your Life
Fabric is a free, open-source AI framework built around a curated library of community-tested prompts ("patterns") for solving everyday problems. This video walks through installing Fabric via the terminal and demonstrates two patterns — extracting wisdom from a video transcript and analyzing claims — using GPT-4. ---
Key Concepts
| Concept | Definition |
|---|---|
| Fabric | Open-source AI framework providing a library of tried-and-true prompts for common tasks; supports GPT-4, Claude, and local/open-source models |
| Pattern | Fabric's term for an individual, granular AI use case — essentially a detailed, well-engineered prompt |
| Stitch | Multiple patterns combined to produce more sophisticated outputs |
| Mill | An optional server component that makes patterns available over a network |
| Loom | A client-side app for calling specific patterns; CLI usage is the primary method shown here |
Notes
What Fabric Does
- Community-generated and reviewed prompts covering a wide range of use cases
- Example use cases:
- Extracting highlights from YouTube videos and podcasts
- Writing essays in your own voice
- Summarizing academic papers
- Generating AI art prompts from written content
- Explaining code
- Converting bad documentation into usable documentation
Installation Steps
Configuration
- Run `fabric --setup` to enter your API key
- Prompts for OpenAI API key; optionally Claude API key
- Setup downloads all patterns automatically
- Restart terminal after setup, then verify with `fabric -`
Using Fabric (CLI)
- List all available patterns: `fabric --list`
- Basic usage: pipe text into fabric with a pattern flag
- Example: `pbpaste | fabric --pattern extract_wisdom`
- Add `--stream` flag to see output in real time as it generates
Demo: extract_wisdom Pattern
- Copied a YouTube video transcript, piped it through `extract_wisdom`
- Output included: summary, key ideas, direct quotes, facts, references, and follow-up recommendations
Demo: Analyze Claims Pattern
- Ran a second pattern to extract and evaluate claims from the same transcript
- Output: individual claims with supporting/refuting evidence and a letter-grade score per claim
Ongoing Development
- New patterns added regularly
- A user interface for easier installation is reportedly in progress
- Works with fast inference providers (e.g., Groq) for near-instant output
Actionable Takeaways
- Clone the Fabric repo and follow the four-step install (clone → cd → poetry → setup script)
- Run `fabric --setup` and enter your OpenAI API key to activate patterns
- Use `fabric --list` to browse available patterns before deciding what to run
- Pipe any text (clipboard, file output, etc.) into `fabric --pattern <name>` to apply a pattern
- Add `--stream` when you want to watch output generate in real time
- If Poetry or shell config causes errors, check your `.zshrc` / `.bashrc` for syntax issues introduced during setup
Quotes Worth Keeping
A pattern is the granular AI use cases, aka the prompts. The stitches are patterns that are put together in different ways to create even more sophisticated outcomes.