capybro.app

AI text rewriting is the process of selecting existing text and having a language model regenerate it — fixing grammar, changing tone, shortening, paraphrasing, or translating — while keeping your original meaning. On Windows, tools like CapyBro turn this into a two-step habit: select text in any app, press a hotkey, and the rewritten version replaces your selection in place. It suits anyone who writes in multiple programs (email, chat, documents, code comments) and wants corrections or tone changes without switching to a separate chat window.

How AI rewriting differs from spell-check and grammar tools

Traditional checkers flag errors against fixed rules and suggest small edits. AI rewriting works at the level of the whole passage:

That difference matters when the problem isn't a typo but that a message sounds too blunt, too long, or too informal for its recipient.

The select → hotkey → replace workflow

CapyBro's model is deliberately minimal, and it works the same way in every app:

1. Select the text you want changed — in Word, Outlook, Telegram, a browser, or an IDE.
2. Press a hotkey (the default shown in the product's own example is Ctrl + Shift + Q).
3. The AI result is pasted back automatically over your selection — the demo on the site shows a rough vacation request turned into a formal leave request in about 0.8 seconds.

Because it operates on whatever text is selected, it doesn't depend on the host app having its own AI feature. The same hotkey works whether you're in a document, a chat window, or a code editor.

Checking the result before it lands

Rewriting in place is fast but risky if the model misreads your intent. CapyBro ships a diff preview (on by default) that shows the original and the AI result side by side before replacement, with options to accept, regenerate, reject, or hand-edit first. Responses also stream in real time, so you can cancel mid-generation instead of waiting for a bad rewrite to finish.

Common use cases

• Fixing grammar and clarity — clean up a draft without rewriting it yourself.
• Changing tone — the site's example converts an informal vacation request into a formal one addressed to a team.
• Paraphrasing — restate a passage in different words.
• Shortening — a "make twice shorter" style command.
• Translating — the interface supports English, Ukrainian, and Russian, while the underlying models can respond in whatever language you request.

You can create custom prompt templates for each of these and assign a separate hotkey to each, so "fix grammar" and "rewrite formally" don't compete for the same shortcut.

Cloud models vs. local AI: the privacy trade-off

CapyBro supports two backends, and the choice is mostly about where your text goes:

• OpenRouter (cloud) — one account gives access to models such as GPT-4o, Claude, and Gemini. Your text is sent to a remote service.
• Ollama (local) — the model runs on your own machine, and in this mode the text never goes to the network at all.

Supporting details from the product: API keys are encrypted locally with Windows DPAPI, and history is stored only on your computer. The free version's code is open source under the MIT license, which lets you verify what the program does — relevant if you handle confidential text.

Choose local (Ollama) if your text is sensitive or you can't send it to a third party. Choose cloud (OpenRouter) if you want stronger models and don't mind the text leaving your machine. If you're unsure, start with local mode and switch per task.

Practical tips for better rewrites

• Give the model a clear instruction, not just "improve this." "Rewrite formally for a client" produces a more predictable result than a vague command.
• Use separate prompts for separate jobs. A dedicated "fix grammar" template won't accidentally change your tone.
• Keep the diff preview on until you trust a given prompt, then decide whether to turn it off for speed.
• Match the backend to the text. Routine grammar fixes can run locally; nuanced tone work may benefit from a larger cloud model.
• Check the output before sending. Rewriting preserves meaning most of the time, but names, numbers, and technical terms are worth a glance.

Availability and cost

CapyBro runs on Windows 10 and 11 (version 2.0.3 at the time of the source material). The free version is described as free forever, and Pro is listed as a one-time $9 payment with no subscription. The source does not specify feature-by-feature limits between the tiers beyond that framing, so check the current download page for what each tier includes before deciding.

OpenRouter is a unified API gateway that gives you access to many AI models from different providers — GPT, Claude, Gemini, and others — through a single account and a single API key. In a text-rewriting tool like CapyBro, it is the cloud option: you paste one OpenRouter key into the app, and the app can then send your selected text to whichever model you choose. The alternative in CapyBro is Ollama, which runs models locally so text never leaves your machine. Choose OpenRouter when you want frontier-model quality and don't mind sending text to a third party; choose Ollama when privacy outweighs model capability.

What OpenRouter actually is

OpenRouter is not a model itself. It is a routing layer that sits between your app and the model providers. Instead of signing up separately with OpenAI, Anthropic, and Google, collecting three API keys, and wiring three integrations, you get one endpoint and one key that can reach all of them.

For a writing tool, that matters because different rewrite tasks suit different models. A short grammar fix is cheap and fast on a small model; a nuanced "convert to formal tone" rewrite may read better from a larger one. OpenRouter lets the app expose that choice without the app developer building a separate integration for every provider.

How CapyBro uses OpenRouter

CapyBro is a Windows 10/11 writing assistant that works in any app with text — Word, Outlook, Telegram, Chrome, VS Code. The flow is: select text → press a hotkey → the text is rewritten, grammar-fixed, or translated in place.

When OpenRouter is the configured provider, that hotkey press sends your selected text through your OpenRouter key to the model you picked, and the result comes back into the app. CapyBro's own description of the feature is that OpenRouter gives you "access GPT, Claude, Gemini in one account." Practical consequences of that design:

• One credential to manage. You enter a single API key rather than one per provider.
• Model choice per prompt. CapyBro supports a separate model per prompt, so "Fix grammar" and "Rewrite formally" can point at different models.
• Streaming and diff preview. Responses stream in as they generate, and by default you see a side-by-side diff of original vs. AI result before it replaces your text — you can accept, regenerate, reject, or hand-edit first.

OpenRouter vs. Ollama in the same app

Both are provider options inside CapyBro, and the trade-off is the same one you'd face in any tool:

CapyBro's privacy notes are explicit about the split: API keys are encrypted locally with Windows DPAPI, history is stored only on your computer, and "in Ollama mode — text never goes to the network at all." That last clause is the tell — in OpenRouter mode, text does go to the network. If your text is confidential, that difference should drive the decision, not model quality.

Typical setup: account, key, paste

The general pattern for connecting any app to OpenRouter:

1. Create an OpenRouter account on their site.
2. Generate an API key in the account dashboard.
3. Paste the key into the app — in CapyBro, into its provider settings.
4. Select a model for the prompts you use.
5. Test with a short, non-sensitive string before using it on real work, and confirm the diff preview shows what you expect.

The expected result after step 3–4 is that your hotkey now returns cloud-model output instead of an error or local-model output. The most common sticking point is a key that was copied incompletely or a model selection left empty — the app has nothing to route to, so the request fails.

Cost and privacy considerations

Two things to check before committing, and neither is something to assume:

• Cost. OpenRouter routes to paid model providers, and CapyBro's own pricing signals are separate from that — CapyBro lists a free version and a one-time $9 Pro payment with no subscription, but that is the app's price, not your model usage cost. Model usage through OpenRouter is a distinct expense. Check OpenRouter's current pricing for the models you intend to use rather than assuming any tier is free.
• Privacy. Anything you send through OpenRouter leaves your machine and reaches third-party providers. CapyBro mitigates the local side (encrypted keys, local history), but it cannot change where cloud inference happens. For confidential text, switch the same app to Ollama mode instead.

A reasonable default: use OpenRouter for everyday writing where quality matters and the text isn't sensitive, and keep Ollama configured for anything you wouldn't paste into a web form.

Crossword Weaver is a Windows-only crossword puzzle maker that supports two distinct puzzle styles: free-form puzzles built from only your own words, and themed symmetrical newspaper-style grids. It offers a demo so you can test it before buying. Choose it if you need both styles in one tool and work on Windows; look at alternatives if you need cross-platform support, a browser-based workflow, or a free tool with no purchase step.

The core difference: two puzzle styles in one tool

Most crossword makers specialize in one output style. Crossword Weaver's distinguishing feature is that it handles both:

If you only ever need one style, a simpler or cheaper tool may cover you. If you switch between casual word-list puzzles and publication-style grids, having both in one program avoids juggling two tools.

Windows compatibility and trying before buying

Crossword Weaver is Windows only. That is a hard constraint, not a preference:

• Windows users: you can run it natively.
• Mac, Linux, or Chromebook users: you would need a Windows environment, or you should choose a cross-platform or web-based alternative instead.
• Before purchasing: the site provides a demo. Use it to confirm the puzzle styles, grid behavior, and output match your needs. The demo is the intended evaluation path, so treat it as your compatibility and feature check rather than assuming the paid version behaves the same in every detail.

The site references purchasing, so plan for a paid product rather than assuming it is free. Pricing details are not specified in the available information, so check the site directly for current terms.

Output: printable and playable puzzles

Crossword Weaver is described as producing printable and playable puzzles. When comparing tools, check these output dimensions on the same basis:

• Print: does the tool export a clean grid plus clues suitable for paper handouts?
• Play: can the puzzle be solved on screen, or is it print-only?
• Format control: can you adjust grid size, clue layout, and numbering?

A tool that prints well but has no on-screen play mode suits classroom handouts. A tool with playable output suits digital assignments or casual solving. Decide which of these you actually need before comparing, because it narrows the field quickly.

Ease of use and automation

Crossword Weaver emphasizes automatic generation: you supply words and clues, and it builds the grid. The practical questions to ask of any maker:

1. How much manual grid adjustment is required after auto-generation?
2. Does it handle word placement conflicts for you, or do you fix them by hand?
3. How long does a typical puzzle take from word list to finished output?

For hobby use, a rougher auto-generated grid is fine. For publishing or repeated classroom use, the amount of manual cleanup matters more than the feature list.

Which tool fits your use case

• Hobby and personal puzzles: a free or low-cost maker is often enough. Crossword Weaver's demo lets you judge whether the two-style support is worth paying for.
• Classroom use: prioritize printable output, fast generation from vocabulary lists, and Windows availability if your school machines run Windows.
• Publishing or newspaper-style grids: the symmetrical themed style is the relevant feature. Compare how each tool handles symmetry and grid polish, since that is where output quality diverges most.
• Cross-platform or browser-based work: Crossword Weaver is not the fit. Choose a web-based maker instead.

How to decide

1. Confirm you are on Windows. If not, stop here and look at cross-platform tools.
2. Decide whether you need both free-form and symmetrical styles, or just one.
3. Download the demo and build one real puzzle you would actually use.
4. Check the output: print it, and try solving it on screen if playable output matters to you.
5. Compare the result against one alternative on the same puzzle, then decide whether the purchase is justified.

The fastest way to choose is to run the same puzzle through Crossword Weaver's demo and one competing tool, then compare grid quality, output options, and time spent.

AI text improvement means selecting text you've already written and having an AI model rewrite it, fix its grammar, change its tone, or translate it — in place, without leaving the app you're working in. CapyBro is a Windows 10/11 assistant that does this through a hotkey workflow: select text, press a key, and the result replaces your selection. It fits anyone who writes in multiple apps and wants corrections without copy-pasting into a separate chat window. The trade-off to decide up front is cloud models (better quality, text leaves your machine) versus a local model via Ollama (text never touches the network).

The core workflow: select → hotkey → result

The mechanism is the same everywhere, which is why it works in Word, Outlook, Telegram, Chrome, VS Code, or any program with a text field:

1. Select the text you want changed.
2. Press the hotkey (the default shown is Ctrl + Shift + Q).
3. The AI processes the selection and pastes the result back automatically — the example on the site converts an informal vacation request into a formal one in about 0.8 seconds.

Because it operates on whatever text is selected in the active window, there's no integration to configure per app. If the app accepts typed text, the workflow applies.

What "improvement" actually covers

Text improvement isn't one operation. CapyBro exposes it as a set of prompt templates, each of which can get its own hotkey:

• Fix grammar — corrects errors without restructuring.
• Rewrite formally — shifts register, as in the vacation-request example.
• Make twice shorter — condenses.
• Translate — the interface supports English, Ukrainian, and Russian, but the underlying models handle many more languages; the response comes back in whatever language you request.

You can create unlimited custom templates, so "improvement" is defined by the prompt you attach to a key rather than a fixed menu.

Cloud vs local: the decision that matters most

This is the choice worth thinking through before you start, because it determines where your text goes.

The site states that in Ollama mode text never goes to the network at all, and that API keys are encrypted locally with Windows DPAPI, with history stored only on your machine. If your text is confidential, local mode is the condition to check first; if you want the strongest models, cloud is the path.

Reviewing before you commit

Replacing text blindly is the main risk of any in-place rewriter. CapyBro's diff preview is on by default: it shows a side-by-side comparison of the original and the AI result before anything is replaced, and you can accept, regenerate, reject, or hand-edit the result first. Responses also stream in real time, so you can read as it generates and cancel mid-stream.

Setup and availability

• Platform: Windows 10 and 11 (version 2.0.3).
• Pricing: A free version exists permanently; Pro is a one-time $9 payment with no subscription. The free version's core is open source under the MIT license.
• Privacy defaults: Keys encrypted locally, history local-only, no ads or tracking.

If you write across many apps and want corrections without breaking focus, the select-and-hotkey model is the reason to use it. If your text can't leave your machine, set up Ollama first and treat cloud models as optional.

CapyBro is an AI writing assistant for Windows 10 and 11 that rewrites, fixes, or translates selected text in place, inside whatever app you're already using. You select text, press a hotkey, and the result replaces your selection automatically — no copy-pasting into a separate chat window. It's a fit if you write a lot across Word, Outlook, Telegram, a browser, or an IDE and want AI edits without leaving the app. It's not a fit if you're on macOS, Linux, or a mobile device, since the current build is Windows-only.

How the select-text → hotkey → in-place rewrite flow works

The core loop has three steps:

1. Select text in any program that accepts text input — Word, Outlook, Telegram, Chrome, VS Code, and similar.
2. Press a hotkey. The default shown in the product demo is Ctrl + Shift + Q, and custom prompts can each get their own hotkey.
3. Get the result in place. CapyBro sends the selection to the chosen AI model and pastes the response back over your selection automatically.

The demo on the site shows a rough vacation request being converted to a formal tone: the original informal message goes in, and a polished version ("Dear team, I would like to request leave from next Monday through Friday inclusive…") comes back and is pasted automatically in about 0.8 seconds.

Two behaviors matter for how it feels in practice:

• Diff preview before applying (on by default) shows a side-by-side comparison of the original versus the AI result before your text is replaced. From there you can accept, regenerate, reject, or hand-edit the result first. This is the main safeguard against an unwanted rewrite silently overwriting your text.
• Streaming responses let you watch the AI type its answer in real time, so you can start reading before generation finishes and cancel mid-stream with one click.

Because the tool operates on whatever text is selected in the active window, it works the same way in every supported app — there's no per-app integration to configure.

Cloud AI vs. local Ollama: how to choose

CapyBro supports two backends, and the choice is mostly about privacy versus model access.

The site states that in Ollama mode, text never goes to the network at all. If your work involves confidential material and you can't send it to a third-party API, local mode is the option that satisfies that constraint. If you want access to frontier cloud models without managing multiple accounts, OpenRouter gives you GPT-4o, Claude, and Gemini under one account.

A practical middle path the product supports: assign a separate model per prompt. That means you can route routine tasks (grammar fixes) to a cheap or local model and reserve a stronger cloud model for tasks that need it.

Setting up custom prompt templates with their own hotkeys

Beyond the default rewrite action, CapyBro lets you create unlimited template commands, each with its own hotkey. Examples given on the site include:

• "Fix grammar"
• "Translate"
• "Rewrite formally"
• "Make twice shorter"

The workflow is: define the prompt once, bind a hotkey to it, then reuse it by selecting text and pressing that key. This turns recurring editing tasks into single keystrokes rather than re-typing instructions each time.

The interface supports three UI languages — English, Ukrainian, and Russian — switchable instantly. The AI models themselves support many more languages; the response comes back in whatever language you request in the prompt.

Privacy and data handling

The privacy-relevant details stated on the site:

• API keys are encrypted locally using Windows DPAPI.
• History is stored only on your computer.
• In Ollama mode, text never goes to the network at all.
• The free version's code is open source on GitHub under the MIT license, so anyone can verify what the program does — which the site frames as especially relevant for confidential text.

If your decision hinges on whether text leaves your machine, the determining factor is which backend you configure: cloud (OpenRouter) sends text to a provider, local (Ollama) does not.

Free version vs. the $9 Pro upgrade

The site describes the free version as "free forever" and Pro as a one-time payment of $9, no subscription. The free version's core is open source under MIT.

The page evidence does not enumerate a full feature-by-feature free-versus-Pro comparison, so the exact boundary between the two tiers isn't fully specified in the available material. What is stated: the free version exists permanently, Pro is a single $9 purchase rather than a recurring charge, and the free version's code is open and verifiable.

Practical notes before you install

• Platform: Windows 10 and 11 only (current version listed as v 2.0.3).
• The hotkey approach assumes text is selectable in the target app — it works in any program with text, per the site.
• Keep diff preview on if you're editing text you can't easily recover, since it lets you review before the replacement is applied.
• Power-user toggles live under Settings → Additional features; most are off by default, so enable only what you need.