Hexis

Visual LLM workflow automation for local files. Connect Input → Prompt → Output nodes on a canvas, write the prompt once, and re-run the pipeline on new documents with a keystroke. Everything stays on your machine.

The workflow, prompts, and inputs are all files in a project folder. Prompts are .prompt.md files versioned alongside your documents. Git-commit them, tweak them, share them – think Makefile for knowledge work.

The bet

Chat interfaces are good at one-off queries, bad at repetition. Notebooks are the opposite. Hexis takes the notebook idea – chain of steps, cached results – and makes each step a node that can accept files, run a prompt, or invoke a Claude Code agent that reads and writes code.

Smart caching only re-runs nodes whose inputs or prompts actually changed. Every workflow run auto-commits its outputs to git, so the history of a document is just the git log. Backends are swappable: Claude CLI, Anthropic API, OpenAI, or Gemini.