2026-07-12

Rowboat: Open-Source AI Desktop for Markdown 2026

Rowboat hit 217 points on Hacker News in July 2026 as an open-source Claude Desktop alternative. It runs locally, respects your privacy, and treats Markdown as a first-class citizen — not an afterthought.

What Is Rowboat?

Rowboat (github.com/rowboatlabs/rowboat) is a desktop AI assistant that runs entirely on your machine. Think Claude Desktop, but open-source, local-first, and designed for writers who work in plain text.

Unlike the hosted Claude or ChatGPT desktop apps, Rowboat connects to whatever LLM you choose — a local model via Ollama, an API key for Anthropic or OpenAI, or a self-hosted inference server. Your conversation history, drafts, and system prompts all live in Markdown files on your filesystem.

Key differentiator: Rowboat stores every draft as a .md file in a user-specified directory. You can edit them with any text editor, version them with git, and pipe them directly into md2rich for publishing.

Why Markdown Writers Should Care

Most AI desktop apps treat output as a black box. You generate text, copy it out, and lose all provenance. Rowboat does the opposite:

Every conversation becomes a Markdown file. Chat logs are saved as structured .md files with YAML frontmatter for metadata — model used, timestamp, token count.
System prompts are Markdown templates. Instead of clicking through a settings panel, you write a natural-language system prompt in Markdown. The file IS the configuration.
Drafts are separate Markdown files. When Rowboat generates a long-form draft, it writes it to a standalone .md file in your drafts folder. You never have to fight a rich-text editor to extract your work.
Git-native workflows. Because everything is plain text, you can commit, diff, roll back, and collaborate on drafts using the same tools your engineering team uses.

Rowboat vs Claude Desktop

Here is how the two compare for a Markdown-first writer:

Feature	Rowboat	Claude Desktop
Source	Open-source (MIT)	Proprietary
Local LLM support	Yes (Ollama, LM Studio)	No (Anthropic API only)
Draft format	Plain `.md` files	App-internal database
System prompt storage	Markdown files in a folder	App preferences (opaque)
Export workflow	Open `.md` → paste into md2rich	Copy-paste from UI
Privacy	Local-first, optional API keys	Cloud-dependent
Cost	Free (open-source)	$20/mo + API usage

The Rowboat → md2rich Pipeline

Here is the end-to-end workflow for a Markdown-first creator using Rowboat and md2rich together:

Draft in Rowboat. Ask the AI assistant to write a LinkedIn article, an X thread, or a Medium post. Rowboat writes the result as a .md file in your drafts folder.
Edit in your text editor. Open the .md file in Obsidian, VS Code, or iA Writer. Review, revise, and add your voice. The Markdown source is yours — no lock-in.
Open md2rich.com. Paste the Markdown into the editor. md2rich converts it to rich text in your browser — zero uploads, zero servers.
Copy and publish. Paste the formatted rich text into LinkedIn, X Articles, Medium, Substack, or Notion. All formatting — headings, bold, lists, tables, code blocks — arrives intact.

Privacy note: Because Rowboat runs locally and md2rich processes everything client-side, your draft never touches a third-party server. This is the difference between writing with an AI assistant and surrendering your writing to one.

Markdown Example: Rowboat Draft → LinkedIn Post

Say you ask Rowboat to draft a LinkedIn post about AI ethics. It writes this Markdown:

# Three AI Ethics Questions Every Team Should Ask

Building responsible AI is not about writing a one-page policy.

It is about embedding ethics into the engineering workflow:

1. **Who audits the training data?**
   - Bias detection should live in CI/CD, not in a quarterly review.
   - Automated fairness checks catch problems before deployment.

2. **What is the fallback for incorrect outputs?**
   - Every API call needs a confidence threshold.
   - Below threshold → human review, not auto-approval.

3. **How do users report problems?**
   - A feedback channel is not a feature request.
   - Users need a direct path to flag harmful outputs.

Now paste that Markdown into md2rich. The output is fully formatted HTML-ready rich text with bold headings, numbered lists, and bullet points — everything LinkedIn expects. Copy it, paste into a LinkedIn post, and publish.

No formatting lost. No third-party platform touching your draft. Just Markdown in, rich text out.

FAQ

Is Rowboat completely free?

Yes. Rowboat is MIT-licensed open source. You pay for compute — either a local LLM (free if you have a capable machine) or API credits for cloud models. The software itself costs nothing.

Can I use Rowboat with a Claude or GPT API key?

Yes. Rowboat supports multiple LLM backends. Configure it with an Anthropic or OpenAI API key, and it works exactly like Claude Desktop — except your drafts live in Markdown files on your disk, not in a proprietary database.

Does Rowboat work offline?

With a local model (via Ollama or LM Studio), yes — Rowboat works fully offline. Your drafts, system prompts, and chat history are all stored locally. No internet required beyond the initial download.

How does md2rich fit into the Rowboat workflow?

Rowboat handles the drafting stage. md2rich handles the publishing stage. Rowboat writes Markdown — md2rich converts it to rich text for social platforms and content management systems. Together they form a complete write → edit → publish pipeline with zero cloud uploads.

What models can Rowboat run locally?

Rowboat works with any model hosted by Ollama (Llama 3, Mistral, Qwen, Phi-3, etc.) or any OpenAI-compatible API server. You can mix local and cloud models per conversation — lightweight edits on a local model, heavy drafting on a larger cloud model.

Bottom Line

Rowboat fills a real gap in the AI desktop market: an open-source, local-first assistant that treats your writing output as yours — not as data trapped in a proprietary UI. Its Markdown-native design makes it a natural partner for any writer who wants to draft with AI and publish from plain text.

The complete workflow is simple: draft in Rowboat, edit in your text editor, convert with md2rich, and publish anywhere. No uploads, no lock-in, no formatting loss.