Starting from a pattern
Building a flow from scratch is straightforward, and it is faster still when you know which pattern you need. Ritemark does not ship ready-made flow templates. What follows are four flows that writers often build, each made from the nodes that come with Flows: a Trigger for inputs, LLM Prompt and Image Generation nodes, Claude Code and Codex nodes for agent work, and Save File for the output. LLM Prompt nodes use your own OpenAI or Google Gemini API key.
Every flow you build is a regular flow file that you can edit freely. Think of these patterns as starting points, not constraints.
Release notes generator
This is one of the most useful flows to build first. If your team maintains a changelog file, whether in KEEP A CHANGELOG format or a simple Markdown list, this flow turns raw developer notes into polished customer-facing release notes.
The Trigger takes your changelog file as a file input. An LLM Prompt node reads it, with instructions to rewrite only the newest entries for a non-technical audience. Technical jargon gets translated into plain language. Internal references get stripped out. Bug fixes get grouped separately from new features so readers can scan for what matters to them. A Save File node writes the result as a Markdown file.
The output is a formatted Markdown file ready to paste into your release page, email newsletter, or product blog. The tone stays consistent because the prompt is the same every time. No more spending twenty minutes per release wondering how to phrase "refactored the event handler debounce logic" for customers who just want to know what changed.
Blog post from outline
Writers who plan with outlines know the gap between having a structure and having a finished draft. This flow bridges that gap. You write an outline with section headings and a few bullet points under each, and the flow expands it into a full first draft.
The Trigger takes your outline file, and two LLM Prompt nodes work on it in turn. The first generates prose for each section based on your bullet points, keeping the structure you defined. The second reviews the complete draft for transitions between sections, making sure the piece reads as a cohesive whole rather than a series of disconnected paragraphs. A Save File node writes the draft to your folder.
You still need to edit the output. The flow produces a first draft, not a final one. But it gets you from outline to editable prose in about a minute instead of an hour. The two-pass approach is important here. A single prompt tends to lose coherence across longer pieces because the model generates each section without seeing what came before. The review pass catches those gaps.
Batch document processor
This flow solves a different problem entirely. Instead of processing one file through multiple steps, it processes multiple files with the same instruction.
Say you have thirty help articles that need their tone updated after a brand voice refresh. Or fifty product descriptions that need a new disclaimer paragraph added. Doing this by hand means opening each file, making the edit, saving, and moving on. Even with AI assistance, the manual file handling adds up.
An LLM Prompt node handles one file at a time, and flows do not loop over a folder on their own. For batch work, use a Claude Code or Codex node instead. Both are agents that work in your folder, so one instruction such as "add the new disclaimer paragraph to every Markdown file in products/" can cover all the files. Put your instructions in the node, name the folder, and review the changed files when the run ends. If you want to check everything before it counts, run it on a copy of the folder or keep the folder in Git.
Give the node enough time for the job. The Claude Code and Codex nodes have a timeout you can raise in the node's settings, and larger batches take longer. These nodes run on your paid Claude plan or ChatGPT account (or an Anthropic API key for Claude Code), so a big batch also uses more of it.
Translation pipeline
The fourth flow tackles multilingual content. If you maintain documentation or marketing content in multiple languages, this flow automates the first-draft translation while preserving your file structure.
The Trigger takes two file inputs: the source file and a glossary of preferred terms. An LLM Prompt node sends both to the model with translation instructions, so brand terms and product names stay consistent. A Save File node writes the translation to the matching place in the other language's folder: if your source is at content/en/getting-started.md, set the output to content/et/getting-started.md.
This is not a replacement for human review of translations, especially for published content. But it eliminates the blank page problem. Your translator or bilingual team member starts with a reasonable draft instead of starting from zero.
Making flows your own
Every flow is a JSON file in your project's .ritemark/flows/ directory. Open any flow in the flow editor and you see all its nodes laid out on the canvas. Change the prompts to match your voice. Swap the file inputs to point at your content. Add nodes if you need extra processing steps. Remove nodes you do not need. If a flow should run on its own, put it on a schedule; it runs while its folder is open in Ritemark.
These patterns are meant to be modified. They show you what a well-structured flow looks like, and then they get out of your way.
Download Ritemark and build the flow that fits your workflow.
Start writing with Ritemark.
Free and open source.