The model will not start
Check that its download is complete, review capacity warnings, and inspect the activity log. A missing cached model needs an explicit download in model preferences.
THE WRITE LOCAL FIELD GUIDE
Get comfortable in your studio. From your first writing sample to your next finished draft, here are the practical parts.
01 · START HERE
The setup wizard saves progress as you continue. Tell Write Local what you write, review your computer’s capacity, choose and start a model, then import a few writing samples and approve a voice guide.
After setup, the app opens to the writing studio. A saved model can start in the background while you read or edit drafts. Change automatic startup in Preferences whenever you like.
02 · YOUR SOURCE LIBRARY
Start with articles, emails, documents, or talks you wrote yourself. Choose files, add a folder, or paste public web links. Review the extracted text before importing it.
Original files remain unchanged. Removing a source invalidates the derived voice guide, so review and rebuild it before generating more work.
03 · YOUR VOICE GUIDE
Choose “Analyze my writing” to find patterns across your library. Analysis reads the full extracted text of every writing sample. Larger collections are processed locally in batches, then the findings are combined into one editable voice guide. Progress shows the analysis and synthesis stages.
If you already have a guide from an earlier version, run analysis again to include your full current collection. The guide is a condensed interpretation of your writing, so review the result.
Edit the guide’s phrasing, add preferences, and approve it. Approval is required before generating drafts. If the guide sounds too formal, too casual, or simply unlike you, change it before continuing.
Useful preference: “Be direct. Use concrete examples. Keep the humor light. Avoid corporate jargon.” Choose instructions that reflect your actual writing.
04 · THE WRITING STUDIO
Choose Chat, Outline, Document, or Presentation. Tell the model the subject, audience, purpose, and constraints. For a presentation, describe the length, topic, and visual direction clearly.
Presentation preview and export share composition, line breaks, and positions. A deck that cannot fit at readable sizes is rejected rather than silently clipped. Make dense slides shorter and move detail into notes.
05 · YOUR LOCAL MODEL
Open Preferences to compare models and review available memory, disk, processor, and graphics information. Capacity estimates are guidance, not measured accuracy or speed guarantees.
Read warnings before loading a model. If the machine feels slow or capacity is tight, choose a smaller model and close demanding applications. Use the short local trial to see how that particular setup behaves.
Automatic startup uses already-downloaded models. The app does not automatically download new weights merely because you opened it.
06 · WHEN SOMETHING NEEDS ATTENTION
Check that its download is complete, review capacity warnings, and inspect the activity log. A missing cached model needs an explicit download in model preferences.
Open its review row and the activity log. Try a text or document export for web pages that require login or heavy client-side rendering.
Make sure a model is running, the voice guide is approved, and the writing prompt is not empty.
Review your sample selection, edit the guide, and give the prompt a clear audience and purpose. The model still needs your editorial judgment.
Shorten the visible points, reduce the detail per slide, and place supporting explanations in speaker notes.
When reporting an issue, include your operating system, app version, steps to reproduce, and a sanitized error. Avoid posting private writing or raw logs containing personal paths.
Get in touchTHE QUESTIONS WE HEAR COMING
The default writing workflow runs a model on your computer through an app-owned Ollama engine. Your imported sources, voice guide, and drafts stay in your local workspace. Model downloads, importing web links, and optional conversion services can use network connections.
A complete desktop package includes its local engine and a guided setup. You choose a model and download its weights.
No. Write Local reads the full extracted text of every writing sample, processes it locally in batches, and combines the findings into one editable voice guide. You review and approve the guide before it steers drafts. It does not retrain the model’s weights.
You can use downloaded models with local writing samples and references offline. Initial model downloads, some first-use document parsing assets, and importing public web pages need a connection.
Import documents, text, Markdown, presentations, and supported public web pages for review. Export written drafts to Word, Markdown, or plain text, and presentations to editable PowerPoint with speaker notes.
It depends on the model and what else is running. Write Local checks memory, disk, CPU, and detectable graphics support, then shows estimates and warnings. Start with a smaller model when capacity is limited; larger is not always better.
The current app is a personal desktop workspace. Shared administration, collaboration, enterprise agreements, and team licensing have not been announced. You can tell us about your workflow through the contact page.
Commercial pricing and purchase terms have not been set. The pricing page will show commercial options when they are available.