The bet behind Fabric is contrarian: the bottleneck in everyday AI use isn't model capability, it's plumbing. People know an LLM could summarize a talk or pull the claims out of an article, but they rarely build the habit because re-wiring the prompt every time is friction. Fabric's answer is to treat each task as a named, reusable 'Pattern' — a Markdown prompt you invoke like a shell command.
What Sets It Apart
- Prompts as files, not chat history. Patterns live as version-controlled Markdown, so the prompt that summarizes a transcript is reviewable, forkable, and shared through a community library instead of locked in someone's chat window.
- Provider-agnostic by design. The same
fabric -p extract_wisdomruns against OpenAI, Anthropic, Gemini, Ollama, Bedrock, and 20+ other backends — changing models is a flag, not a rewrite. - Built for the pipe. Output is plain text meant to flow into
grep,jq, or the next Pattern, so AI becomes one more composable Unix-style tool rather than a separate app you context-switch into. - More than a CLI. A web UI and REST API expose the same Patterns for non-terminal users and server deployments.
Who It's For
Great fit if you live in the terminal, want your AI workflows under version control, or prefer a shared prompt library over reinventing prompts from scratch. The Go rewrite ships as a single fast binary, with no Python environment to manage. Look elsewhere if you want a polished consumer chat app with memory and account sync — Fabric is deliberately a tool, not a product, and its payoff scales with your willingness to wire Patterns into your own scripts.
