A JournalLM Alternative That Runs in Your Browser
JournalLM popularized the private, offline AI journal as a desktop download. Writed brings the same idea to the browser — nothing to install, works on any OS.
JournalLM helped make a great idea mainstream: a journal where an AI reads your entries and talks back, with everything kept private and offline on your own machine. If that's what you're after, you have options — and one of them needs nothing more than a browser tab.
The same core promise
Both JournalLM and Writed are built on the same principle: your journal is yours, the AI runs locally, and your words don't get uploaded to a company's servers for processing. Neither one is a cloud chatbot pretending to be private.
The difference is how they run.
Desktop app vs. browser
JournalLM is a native app you download and install on macOS, Windows, or Linux. Writed is a website. You open a URL and start writing — there's nothing to install, update, or grant system permissions to.
| JournalLM | Writed | |
|---|---|---|
| Install | Download per OS | None — open a link |
| Platforms | macOS / Windows / Linux | Any modern browser |
| AI location | On your device | On your device (in the tab) |
| Source | — | Open source (MIT) |
| Price | — | Free |
The browser approach has real upsides: it works the same on any operating system, you can try it instantly, and because it's a static site there's no auto-updater or background process. The trade-off is that browser AI leans on WebGPU, which is strongest on desktop — so Writed is desktop-first, with a slower CPU fallback on phones.
How the AI actually stays private
When you save an entry, Writed breaks it into chunks and computes a vector embedding for each one using a small model (bge-small-en) running in your browser. Ask a question and it retrieves the most relevant chunks with hybrid search — semantic similarity, keyword matching (BM25), and a recency boost — then a chat model (Llama 3.2 3B on WebGPU, or Qwen 2.5 1.5B on CPU) answers using only your own entries as context.
All of that happens locally. The only network request Writed ever makes is the one-time download of the model files on first visit, from public CDNs like Hugging Face and jsDelivr. Those CDNs see your IP — the same as loading any web page — and never see your journal.
No lock-in, ever
Your entries are plain Markdown. Export the whole journal as a zip of .md files with YAML frontmatter at any time, and open them in Obsidian, Logseq, or a plain text editor. Because Writed is open source, you can also read exactly how it works.
Try Writed now
No download, no account. Open it, write an entry, and ask it something.
Frequently asked questions
- Is a browser AI journal as private as a desktop app?
- Functionally yes — in Writed the embedding and chat models run inside your browser tab, and entries are stored in the browser's local IndexedDB. Nothing is uploaded after the one-time model download. You can verify it in the Network tab of DevTools.
- Do I need a powerful computer?
- A modern laptop or desktop with WebGPU runs the full chat model smoothly. CPU-only and mobile devices fall back to a smaller, slower model. Desktop is recommended.
- Does Writed work offline?
- Yes. After the first visit downloads the model files, Writed runs fully offline — the page itself is a static asset.