From first draft
to finished piece.

Get started

Write your article on the page. Explore ideas in frames. Bring back only the work you choose.

Writer is a local macOS writing app. Start in Write for focused drafting or Augmented for working with an AI agent beside your document. Turn on the optional Website Editor in Settings to review and edit local page drafts. You can create, edit and save documents without connecting an AI provider.

1. Create or open a document using the icons at the upper left.
2. Write a few paragraphs, or import a Word document.
3. Connect a provider when you want AI assistance.
4. Switch to Augmented, send a section to a frame, and ask for a specific improvement.
5. Review the frame’s output, then explicitly send the content you want back to the editor.
6. Keep writing locally or export a Word copy.

Installation

The current packaged build is for Apple Silicon Macs running macOS 13.5 or later. Check the Writer download page for the current alpha, version, checksum, and installation help. The download becomes available after the installer is published.

1. Download the Writer .dmg file.
2. Double-click it to mount the disk image.
3. Copy Writer.app into your Applications folder.
4. Eject the disk image, then launch the copied app from Applications.

The app bundles its agent runtimes. You do not need to install Node, Bun, Codex or Claude Code separately. Internet access is needed for provider sign-in and online AI requests; local writing does not require it.

Updating Writer

Quit Writer after your document has saved, then replace the application with the newer supplied version. Your documents are stored outside the app. Keep a backup of their folders, including hidden .writer data.

Create and open documents

Create a new document

1. Click the new-document icon at the upper left, or press ⌘ N.
2. Choose the folder and filename in the native dialog.
3. Confirm, then begin writing.

Choose a location you can find again, such as a folder for the article or project. Writer saves your writing to disk automatically.

Open existing work

Click the folder icon or press ⌘ O, then choose Choose a file. Open a Markdown, HTML or Word document. Opening a .docx imports an editable copy; it does not turn the Word original into a file that Writer edits in place.

Import an existing blog article

1. Open the folder menu and choose Import a blog from URL.
2. Enter the full public article address, including its domain, and choose Import blog.
3. Wait while Writer imports the article and supported images into your local folder.
4. Review the imported text, source attribution and any notices before rewriting it.

This makes a local article you can edit in Write or Augmented. Use Website mode when you want to work with a page’s design and metadata. Importing needs internet access but does not start an AI request. Pages that require sign-in or rely on scripts may not import completely.

Save and recover

Look for the save status before quitting or changing documents. Press ⌘ S to save pending edits immediately. If saving fails, keep the window open and resolve the reported problem, such as an unavailable folder. Autosave is not a versioned backup service.

Keep the document’s folder and its hidden .writer folder together when backing up or moving a project. Canvas state, managed references and local media depend on that supporting data. Copying only the article file may omit those materials.

Find commands

Press ⌘ K to open the command palette. Search for a command or a document. It also provides Word import/export and writing preferences.

Write without distractions

Choose Write or press ⌘ 1. The canvas disappears and the document becomes the focus. Move the pointer to the top edge to reveal the toolbar when it is hidden.

Use the command palette to toggle typewriter scrolling, which keeps the active writing area in view, or focus dimming, which reduces emphasis on surrounding text.

Spelling

Writer enables macOS spell checking in the editor. Misspelled words can receive a red underline. Right-click a marked word and choose a suggested correction. Suggestions depend on the system’s language and spelling settings.

Move or remove a paragraph or section

1. Move over the block and use its handle to select it.
2. Drag the handle to the desired position in the document, following the insertion indicator.
3. Release to move the selected block.

With a whole block selected through its handle, Delete or Backspace removes that block. Use ⌘ Z to undo. Clicking into text resumes ordinary text editing.

Use Augmented mode

Choose Augmented or press ⌘ 2. The editor and canvas appear side by side. A frame is an independent place to explore a rewrite, an image or a different presentation of your material.

Start from your document

1. Select a section or paragraph using its handle in the editor. The arrow toward the canvas activates only while a section is selected; clicking outside, moving the caret or pressing Escape clears that selection.
2. Use the rail arrow pointing toward the canvas to send it across. With a maximized frame, the action can target that frame; otherwise follow the new highlighted frame.
3. For several sections, use the rail’s section chooser. Select the sections you want, or choose the entire document.
4. Enter your request in the frame’s input and send it.

Sending content across does not itself start an AI request. New frames are highlighted and brought into view so you can see where the content landed.

Start fresh

Click + New frame to create a blank workspace. You can also use the canvas input to begin new work. Give the agent a clear task, audience and desired output.

Review and bring work back

The frame preview shows its current output even when the card is not maximized. Expand the frame to work in more detail. Read the result, place the cursor where you want it in the editor, and use the rail arrow toward the editor. Review what you insert and remove superseded text when needed; inserting a frame is not a promise to replace the original passage automatically.

Work on several frames at once

Start a request in one frame, then submit a different task in another. Frames can work in parallel. Each captures its own model and reference selection when the request starts. Use a frame’s stop control to cancel its active request.

Processing updates appear inside the frame conversation. Expand or minimize those updates to see more or less of the activity without creating a separate panel.

Arrange your canvas

• Rename: click a frame’s title and edit its canonical name.
• Move: drag a frame by its header.
• Expand: use the frame’s expand control to focus on its output and conversation.
• Delete: use the trash icon on the frame. Deleting a frame is separate from removing content already inserted into the editor.
• Zoom: use the plus and minus controls to change scale. Double-click empty canvas to fit the current arrangement. At 100%, a large collection can extend beyond the visible area; pan to explore it.
• Arrange and fit: use the square overview control to place visible frames in a non-overlapping arrangement and bring them into view. This changes their positions; Writer saves the arrangement with the document.
• Swap sides: use the swap icon at the top of the middle rail. Writer remembers which side holds the canvas.
• Resize: drag the divider; double-click it to restore equal widths.

The canvas starts at 100% zoom on app startup and for a new document. Fitting all frames may use a smaller scale.

Work on website pages

Use the page design, editor and canvas to improve a website while keeping changes local. Website Editor is off by default. Open Settings and turn on Website Editor, then choose Website beside Write and Augmented. The setting persists across restarts. Turning it off saves the active draft first and keeps your pages and connection.

Open or create a page

Choose New page to start with an editable blank page, enter a public URL and choose Import page, or choose Import HTML. The page selector opens saved pages. In Website mode, the New and Open icons, ⌘ N and ⌘ O open this page chooser without switching to Writer. An invalid address shows an error beside the input; correct it and try again.

Opening and importing show progress until the page is ready to use. Imports automatically save supported styles, fonts and images. Saved pages reuse these assets and attempt to repair missing resources once when opened. If resources are still missing, open Page actions → Page information → Retry page assets. Page information also lists unsupported resources.

Choose Open from Robynn for connected pages and saved cloud audits; see Open pages and audits from Robynn. Local page creation and HTML import work without a Robynn account.

Edit and preview

Pages open in Design. Click a mapped heading, paragraph or list item to edit its text in place; Enter finishes the edit. Use View → Text to edit the same draft as a document. Undo or ⌘ Z reverses page changes, including a section replacement. While actively typing, the editor keeps its normal text-undo behavior. Website change history is kept for the current page session, with a bounded number of changes.

The View menu contains Desktop, Tablet and Mobile widths, Actual size, Fit to pane, and the original/draft comparison. Drag the divider to give the editor or canvas more room; only a small strip remains so you can reach the other pane. Preview shows the draft across the workspace, and Back to editing restores the split. Changes save automatically, including text you are still editing in Design. Saved locally confirms completion. ⌘ S can also flush pending changes immediately. A failed save stays visible with a retry action.

Edit page metadata

1. Open the Page actions menu (•••), then choose Page information….
2. Set the SEO title, description, canonical URL and robots instructions. Open Open Graph for a sharing title, description, image address, image description, site name, URL, type and locale.
3. Use Additional metadata for author and keywords. Check the Google search and Open Graph previews beside the form.
4. Choose Save page details and wait for Saved locally. Reset returns unsaved fields to the draft’s saved values.

Metadata changes stay in the local page draft, together with its design and content. The previews estimate how a search result or shared link may look; search engines and sharing services can choose different text, crop images or show an older cached version. An image preview may need internet access to load its public image address. Setting noindex shows a reminder that the page asks search engines not to index it.

Review suggested changes

1. Open the Review tab, choose Content or SEO / GEO, then choose Find improvements. This uses your configured AI provider. Importing, connecting and choosing references do not start an AI request.
2. Review each suggestion, its reason and the current text. Hover over a suggestion to locate its section on the page.
3. Preview the change, then apply or dismiss it. Changes to the draft or context can make earlier suggestions stale.
4. Wait for Saved locally, then reopen the page from the page selector to continue.

Work on a section in Canvas

1. Open Canvas and choose Select section. Click a div or section in the page; use Select parent when you need a larger section.
2. Use the middle rail's send arrow to create a frame. The frame shows the section's visual preview with its saved page styles.
3. Ask the agent to change that section, such as revising its layout or adding an image. Review the original and modified previews.
4. Select the changed frame and use the return arrow to replace the original section. If the page changed in the meantime, select the section again to avoid replacing newer work.

New pages start with an empty canvas. Review analysis stays in the Review tab. Canvas reuses prepared section previews when you return to it; progress remains visible until a section is ready. Each page keeps its own draft, frames and references. Frame output is applied only when you choose to send it back.

Open pages and audits from Robynn

This connected library needs a compatible Robynn service and access to a workspace and website. An authorization page in your browser is only one step: return to Writer and wait for Connected and the website selector.

1. Open the page selector, choose Open from Robynn, then connect or choose your website.
2. In Pages, search by title or URL. Select one or several pages and choose Download selected. Use More pages when available.
3. Watch each page’s download result. Choose Open downloaded to work locally; retry only pages marked as failed. Pages marked Fetch from website retrieve their public URL when you choose to open or download them. Other unavailable pages show their reason.
4. Choose Cloud audit to read the website’s saved report, scores, findings and reported pages. Search findings or select one to inspect its recommendation and evidence. Refresh report fetches the saved report; it does not start another audit.

Downloaded pages keep separate local drafts, canvases and references. Loading a website or its report does not start an AI request. When brand context is available for a connected page, expand Brand context, load it and select the references to save with that page. Only that selection is included in the page analysis.

Keep control of the source

Opening the same remote revision returns to its local draft. If the source changes, Writer offers the existing draft or a new snapshot. Disconnect removes native connection credentials from Keychain; saved pages and selected references remain local. Remote revocation failures are reported separately.

Prepare articles for publishing

The Publish control beside an open document’s title starts a reviewed article workflow with Robynn. It is available from Write and Augmented; you do not need to enable Website Editor.

Prepare and review a preview

1. Open the article, choose Publish, then Connect Robynn for publishing if requested. Complete the browser flow and return to Writer.
2. Choose the connected Website. Set the article title, author, template, URL name, excerpt, categories and tags offered by that website.
3. Choose Prepare preview. This saves the current article and sends that article and its included images to the selected Robynn workspace. Reference-file collections, unrelated documents and frame conversations are not part of the publication package.
4. Inspect the private website preview at desktop and mobile widths. Read any notes about how the article fits the template.
5. Choose a reviewer and Send for review. The assigned reviewer can approve that revision or request changes when their permissions allow it.
6. After approval, choose Publish article to send the approved revision to the live website. Review and publication are separate actions.

When the local article has newer edits, use Prepare updated preview. The revised content needs its own review before publication. Autosave does not upload or publish the article.

Read the publication result

WordPress accepted this change means the destination accepted the write. Published and verified means the subsequent verification completed. If verification fails, the change may already be live: use Check status or Recover publication before trying another publication. If the website changed after your preview, review its newer revision before preparing another change.

Images sent to WordPress may become publicly accessible before the post is published. Inspect the included images and destination before proceeding.

Follow up or recover

When offered by your workspace, Check this article again in seven days asks Robynn to check the page and available measurements even while Writer is closed. It does not edit or republish the article. Prepare previous version and Prepare unpublish, when available, create a separate revision for fresh review and approval.

For company writing, the optional Company article workflow in Document context can make the selected website’s knowledge, article inventory and saved audit available to the agent. You still choose the content, reviewer and publication action.

Manage context and reference files

Writer distinguishes the article you are writing from supporting files and the work in each frame.

Add shared document context

1. Click the paperclip immediately after the Open folder icon in the upper-left toolbar.
2. Drop several files into the Document context panel, or use its file selector.
3. Wait for each file’s reading status. Click a filename to preview its image or extracted text.
4. Choose Done, press Escape, or click outside to close the panel. The paperclip count remains visible when the toolbar is shown.

Every frame uses all document references by default, including files added later. Open the frame’s source selector to choose a saved subset. A different document has its own context. The files and their assignments survive app restart. The paperclip also works for the active page in Website mode.

Choose company context for this document

Company knowledge appears below Reference files. Connect to a Robynn organization before enabling Company’s Robynn Growth Cortex. This lets the agent search the organization’s available company knowledge for new requests. The option is specific to this document; local reference files remain available without it.

For a company article, choose a website and enable Company article workflow when your workspace supports it. This adds access to the selected website’s company knowledge, article inventory and saved audit under your organization’s provider permissions. Turning on context does not run the agent or publish the article. If saved company context becomes unavailable after disconnecting, use its remove control to continue with local sources.

Company knowledge and the article workflow require the matching Robynn cloud services. A saved selection alone does not prove that the remote source is available; follow the connection and request status.

Add a reference to just one frame

Use the paperclip in that frame’s input area. Attach the reference there rather than in Document context when it should apply only to that task.

Continue the same conversation

Each frame keeps its conversation across messages, document edits and app restarts. Earlier decisions remain part of the thread while new requests receive the latest article and frame. Progress appears while the provider restores or summarizes a long conversation. Changing providers restores saved chat into a new provider session; earlier tool results remain in the previous provider’s log.

Each sent message includes a source receipt showing the files, versions and extraction status prepared for that request. This records available context, not proof that every passage was read.

Formats and limits

Supported files: PNG, JPEG, GIF, WebP, PDF, DOCX, TXT, MD, CSV and JSON. Each context accepts up to 12 files, 10 MB per file. Text previews show up to 8,000 characters. Scanned or textless documents may not yield usable extracted text; follow the reported status instead of assuming they were fully read.

Replace, remove and troubleshoot

Remove a reference with its remove control and attach a newer copy when needed. Writer keeps managed copies, so changing the original elsewhere does not automatically refresh the attachment. The original file is never changed by removal. Use Retry for a reported loading or extraction failure. Sending waits for extraction to finish. A failed or unreadable source must be retried, removed, or explicitly excluded before sending; your unsent message stays in the input. XLSX workbooks need to be exported to a supported format first.

Removal is held while an affected request is running. New attachments apply to subsequent requests. Each new request captures the current reference list along with the latest article and frame. Bounded excerpts are provided, and the agent can read more through reference tools; attaching a large file does not mean every word is placed into every prompt.

Connect AI providers

Open the gear icon or press ⌘ , to configure agent providers in the right-side Settings drawer. Your document remains visible. Press Escape, click outside, or click the gear again to close it.

Sign in with an account

1. Choose Sign in with account on a supported provider card.
2. Complete the browser sign-in flow shown by the provider.
3. Return to Writer and check that the card reports a connected account.
4. Choose an available default model.

Writer’s account connections are kept separately from terminal sessions for Codex, Claude and Gemini. Signing out in Writer does not sign out those terminal sessions.

Use the Google account option

Google Gemini has separate Code Assist account and API key sections. Writer’s current account setup is intended for eligible Gemini Code Assist organization accounts. Check Google’s current authentication guidance for your account type and use the route available to your account; a Google subscription alone is not confirmation that it works in this Writer build.

1. In the Code Assist account section, choose Sign in with Google. If your organization requires a project, open Cloud project (optional) and enter the project ID it provides.
2. Finish sign-in in the browser and return to Writer. Saved sign-in means the connection is stored on this Mac; access is checked when you use it.
3. Choose Load models or Refresh to fetch the account’s model list, then select an available default. Use Retry models if loading fails.

Changing the Cloud project requires reconnecting. Account and API-key profiles stay separate, so existing API-key frames retain their route. Account availability and quota are determined by Google. This account route does not generate images.

Connect an API key

1. Choose Use API key.
2. Enter the key only into the provider’s key field, then save it.
3. Check the connection result and choose a model.

On the OpenAI and Anthropic cards, saving a key switches new prompts to that key while retaining the saved account connection; check the active route before sending. Google keeps its account and API-key profiles separate. Do not put keys in documents, reference files, skills or chat prompts. API-key usage and subscription-account usage are different connection routes. Consult your provider account for eligibility, billing and usage limits.

Choose a model per frame

Click the model name in a frame’s input area. Choose from the connected providers and their available models. Writer saves the choice with that frame, so two frames can use different models. The canvas input also lets you select a model for new work.

A frame’s choice does not change the default in Settings. The picker is disabled while that frame runs. If a saved model or connection is no longer available, choose another; Writer should report the problem rather than silently substitute a provider.

Generate and use images

1. Open or create the frame where you want the visual.
2. Attach a reference image if it helps, and describe the subject, purpose and desired composition.
3. Ask for generation. For example: “Create a landscape editorial illustration for the current article about secure data handling. Avoid embedded text.”
4. Review the resulting image in the frame before bringing the output into the editor.

Writer supports image generation through a connected ChatGPT/Codex account or an OpenAI API key. Check the OpenAI settings card for the active image-generation route. Account generation uses Codex usage limits; the API route uses its key. Successful sign-in alone does not establish that a particular request is within your provider’s current limits.

If generation fails, inspect the error, verify the OpenAI connection and selected model, and retry after resolving the reported issue. A text-provider connection alone does not guarantee an image-generation connection.

Format your document

Use structure to make the article easy to scan. Writer supports headings, paragraphs, lists, links, tables, blockquotes, code, images and supported cards. It is a writing environment, not a full page-layout application.

Format selected text

Select text, right-click it, and choose Heading 1 through Heading 4 or Body from Format as. The choice changes the selected paragraphs’ structure. Use ⌘ Z to undo it. Right-clicking at a caret without a text selection retains the native menu, including spelling suggestions.

While typing

• At the start of a paragraph, type # followed by a space for a heading. Use two through four hashes for smaller heading levels.
• Type -, + or * followed by a space to start a bullet list.
• Inside a list, Enter starts another item. Tab indents and Shift–Tab outdents where supported.
• Use Tab to move between table cells when you are not editing a nested list.
• Use Undo if an automatic formatting change was unintended.

Ask for a different presentation

Send the relevant passage to a frame and request a supported structure:

Review both wording and structure before inserting the result. Arbitrary frontend components, JavaScript and unrestricted HTML/CSS are not guaranteed to render. A design skill can guide the agent toward supported formats; it cannot expand the editor’s rendering capabilities by itself.

Import and export Word documents

Import a Word document

1. Choose File → Import Word Document, use Open Document, or find the import command with ⌘ K.
2. Select a .docx file.
3. Writer creates an editable Markdown copy in the current folder and preserves the original source separately.
4. Review the imported content and any simplification notice before editing.

Export to Word

1. Open the article you want to export.
2. Choose File → Export as Word Document, or search for the export command in the palette.
3. Choose a filename and destination in the Save dialog.
4. Open the exported file in Word and review it before sharing.

Export saves current edits first. It creates an explicit Word copy; subsequent autosaves continue updating Writer’s document, not that exported copy. Export again when you want a newer Word version.

Legacy .doc files are unsupported. Word archives are limited to 32 MB. Remote images are not automatically downloaded during export; unsupported images can produce a notice or placeholder. This release uses one export style rather than selectable Word design presets.

Use writing skills and extensions

Skills: reusable writing guidance

A skill is a Markdown file containing instructions, such as an editorial style or document-design approach.

1. Open Settings and find Skills & Extensions.
2. Open the skills folder: ~/.writer/addons/skills/.
3. Drop in a Markdown file, or a folder containing SKILL.md.
4. Refresh the list to check for diagnostics. Restart Writer if needed, then ask a frame to use the skill by name.

Edits to valid skill files are read for new requests. Removing a file removes its instructions from subsequent prompts. Skills apply across documents, so use document context for material intended only for one article.

---
name: Clear editorial style
description: Make explanatory writing easier to read.
---
Preserve the facts in the current article.
Prefer short paragraphs and descriptive headings.
Define unfamiliar terms. Do not invent quotations.

Each skill must be UTF-8 and no larger than 32 KiB. Writer loads up to 32 skills within a 128 KiB total instruction budget. Supporting files beside SKILL.md are not automatically loaded or executed.

Extensions: additional agent tools

Copy a compatible extension’s complete package folder into ~/.writer/addons/extensions/, then quit and restart Writer. A package includes manifest.json and a JavaScript entry file. Refreshing the list checks discovery; restart ensures agent processes load the installed version.

Compatible extensions do not require rebuilding Writer. To remove one, remove its folder and restart. Extensions execute local code with your user account’s filesystem and network access, so install only code you trust. Writer does not install their dependencies for you.

For authors, see the companion extension contract and skill format documentation in this documentation folder.

Keyboard shortcuts

Hover over rail controls to see the shortcut for the current panel arrangement. Swapping sides changes the direction of transfer. Keyboard behavior can depend on which editor, input or dialog currently has focus.

Troubleshooting and good habits

Before sharing a finished piece

• Check facts, quotations and the source of important claims.
• Confirm only the intended frame outputs were inserted.
• Review image placement, captions, links and tables.
• Check save status and make a backup of the project folder.
• Open the exported Word copy and inspect it.

Online AI requests send relevant prompts and context to the selected provider. Local storage does not make those requests offline. Share only material you intend that provider to process.