Fabric

---

title: fabric radar_quadrant: Tools radar_ring: Assess radar_position: inner created: 2026-05-22 last_updated: 2026-05-22 related: ["STORM", "Hammerspoon", "LLMMaintainedWiki"]

fabric

fabric is a CLI framework for applying AI prompt patterns to any text input via Unix pipes. It ships with a crowdsourced library of Markdown-based "patterns" — each a focused system prompt for a recurring task — and supports 20+ LLM providers including local models via Ollama.

How It Works

Patterns are Markdown files stored in a directory. Each pattern contains a system prompt for a specific task. Pipe any text through a pattern:

# Summarise clipboard content
pbpaste | fabric --pattern summarize

# Extract key insights from a YouTube transcript
yt --transcript <youtube-url> | fabric --pattern extract_wisdom

# Find security threats in code
cat app.py | fabric --pattern find_threats

# Generate tags for an article
cat post.md | fabric --pattern create_tags

Patterns can be chained ("stitches") using standard Unix pipes:

yt --transcript <url> | fabric --pattern extract_wisdom | fabric --pattern create_quiz

Pattern Library

The built-in library includes patterns for: summarisation, wisdom extraction, quiz generation, threat analysis, content rating, essay improvement, code review, meeting notes, PR description, and many more. Custom patterns live in a private directory and are never shared.

Provider Support

20+ providers via a --model flag: OpenAI, Claude (Anthropic), Gemini, Ollama (local), Azure OpenAI, AWS Bedrock, and others. Switch models per invocation without changing the pattern.

Installation

curl -fsSL https://raw.githubusercontent.com/danielmiessler/fabric/main/scripts/installer/install.sh | bash
# or
brew install fabric

Also available as a Docker container and REST API server with streaming.

Positioning

fabric occupies the "prompt reuse" layer — named, version-controllable patterns replacing ad-hoc LLM queries for recurring tasks. Complementary to STORM (which generates long-form research from scratch) and llm CLI (which handles one-off queries). The patterns-as-files model makes the library forkable, auditable, and extensible without coupling to a specific LLM provider.

Radar Assessment

fabric sits at Tools → Assess inner. The prompt reuse problem is real for anyone who regularly applies LLMs to articles, transcripts, or code reviews. The Unix pipe design composes with existing workflows without introducing new orchestration. Ollama support enables fully local, privacy-preserving operation. Trial gate: at least three recurring LLM tasks replaced by fabric patterns used consistently over two weeks.