After 4 months solo: I built CapyBro — an open-source AI text helper for Windows
Hey IH,
After 4 months of nights and weekends, I shipped CapyBro — a Windows tray utility that brings AI to any text via a global hotkey.
What it does
You select text anywhere in Windows (Word, Telegram, Outlook, Chrome, VS Code — anywhere). You press Ctrl+Shift+E. AI rewrites it in place. That's the whole product.
Use cases that actually matter to me:
Fix grammar in messages before sending
- Translate paragraphs from Ukrainian to formal English for client emails
- Paraphrase paragraphs without retyping
- Run my own custom prompts ("rewrite as a technical bullet point")
Why I built it
I was paying $20/mo for ChatGPT Plus and using it 30+ times a day for tiny text fixes. The friction of Alt+Tab → ChatGPT tab → paste → wait → copy → paste back was killing my flow. So I built the simplest possible thing: AI as a system-wide hotkey.
Tech & decisions
.NET 8 + WPF (not Electron). 48 MB installer. <1s cold start. Native feels right for a system utility.
- MIT license for the core. I almost went "source-available" but decided trust > control. Forks are fine.
- Pro $19 one-time, NEVER subscription. Hard line for me. I personally hate productivity-tool subscriptions.
- Two AI providers: OpenRouter (cloud, one key for many models) + Ollama (local). The local mode is what lets me work on confidential client docs without paranoia.
- 3 UI languages (EN/UK/RU). Most "AI text tools" assume you write in English only. Wrong.
What was hardest
The Win32 paste-back logic. Specifically, restoring foreground + keyboard focus to the right window (and the right child control inside it — like the Scintilla edit field inside Notepad++) after a modal dialog steals focus. Took 3 iterations and a lot of GetGUIThreadInfo + AttachThreadInput sandwiches.
Second hardest: trimming the installer. .NET 8 self-contained is ~150 MB initially. NSIS lzma /SOLID compression got it to 48 MB.
Marketing so far
Zero. I literally just shipped. Posting here is part of the launch plan. Next steps:
Product Hunt next week
- Show HN
- AlternativeTo listing (in cooldown until June 4)
- Comparison content on capybro.app/vs/
What I'd love feedback on
Is the workflow obvious from the landing page?
- Anyone building similar things — what'd you do differently?
- Pro tier pricing: $19 one-time. Too cheap? Too expensive?
Site: https://capybro.app
GitHub: https://github.com/phantasmat2018/capy-bro
Direct download: https://github.com/phantasmat2018/capy-bro/releases/tag/v2.0.0 (Windows 10/11 x64, 48 MB)
Thanks for reading 🦫
— Roman
🚀 LIVE NOW on Product Hunt!
https://www.producthunt.com/products/capybro
If anyone here has 30 seconds, would massively appreciate your honest take. Will be replying to every comment for the next 8 hours.
Update — launching on Product Hunt tomorrow morning! (Thursday 5 June, 10am Kyiv / 12:01am PDT)
Pre-launch page is live now:
https://www.producthunt.com/products/capybro
For anyone curious about the technical side: I wrote a deep-dive on dev.to covering the Win32 paste-back rabbit hole that took the most iterations:
https://dev.to/phantasmat2018/after-4-months-solo-shipping-a-windows-tray-ai-hotkey-on-net-8-wpf-and-the-win32-paste-back-50hm
Would love your support tomorrow — and if you're an IH'er who launches on PH, I'd love any 2026-era tactics you've learned (the playbook seems to shift every quarter).
Update: I also wrote a technical deep-dive on dev.to covering the
Win32 paste-back rabbit hole with real code from the repo:
https://dev.to/phantasmat2018/after-4-months-solo-shipping-a-windows-tray-ai-hotkey-on-net-8-wpf-and-the-win32-paste-back-50hm
Sharing here in case anyone wants the technical side of how the
tray utility actually works under the hood.
This is a very clear workflow, and that is the strongest part. The value is not “another AI writing tool.” It is removing the tiny repeated friction of moving text between apps, ChatGPT, and the original window.
The system-wide hotkey angle is strong because it makes AI feel like part of Windows instead of another tab. The local Ollama option also gives it a better trust story for client docs, which matters a lot more than most productivity tools admit.
One thing I’d pressure-test before Product Hunt and Show HN is the brand frame. CapyBro is memorable and friendly, but the product itself feels more serious than the name suggests. You are building a native AI text workflow layer for Windows: rewrite, translate, paraphrase, custom prompts, local/private mode, and fast paste-back across apps.
Xevoa .com would fit that broader direction better if you want the product to feel like a clean productivity/workflow brand rather than a small utility. The product can still stay simple, but the name should give it room if it expands into more actions, teams, enterprise users, or deeper Windows automation.
Since you are about to push PH, HN, AlternativeTo, and comparison pages, this is probably the right time to pressure-test the name before more public assets and user memory lock around CapyBro.
Thanks Aryan, this is genuinely useful pressure-test thinking.
The brand name was a deliberate choice — capybara is friendly, memorable, and signals "this is an indie tool, not enterprise SaaS." That maps to my actual differentiation (open source, $19 one-time vs Grammarly's $20/mo subscription, indie-built). A more sterile name would actually weaken that positioning.
You're right that I should pressure-test before bigger launches. Plan: I'll watch the next 2 weeks of feedback (PH launch, HN, DOU publication) and see if "name doesn't fit" comes up repeatedly. If 5+ people independently say it — that's signal worth acting on. A single comment is noise, even a thoughtful one.
For context on the seriousness fear: my technical deep-dive on dev.to went live yesterday with the same brand name and got solid reception from the .NET community — https://dev.to/phantasmat2018/after-4-months-solo-shipping-a-windows-tray-ai-hotkey-on-net-8-wpf-and-the-win32-paste-back-50hm
But I'll keep listening 🙏
That’s fair, and I actually think your reasoning makes sense if the indie/open-source/friendly angle is intentional.
In that case, I would not judge CapyBro like an enterprise SaaS name. I’d judge whether it helps the exact launch story you want:
friendly native Windows AI tool
one-time price
open-source trust
local Ollama option
faster than jumping between tabs
That can work if the positioning makes the friendliness feel deliberate, not accidental.
The bigger thing I’d watch before PH/HN is not only whether people mention the name. It’s whether they instantly understand the category from the first line. If they remember CapyBro but explain it as “some cute AI writing app,” the positioning is leaking. If they explain it as “the Windows hotkey AI tool that rewrites text anywhere,” you’re fine.
Drop your email and I’ll send over a tighter launch-positioning version. It’ll be easier to make useful in writing than turning the thread into a full teardown.
The first-line category test is the right one — I'm stealing it as my actual metric. "The Windows hotkey that rewrites text anywhere" vs "cute AI writing app" is exactly the line I want people to land on, and you're right the name can pull it toward the second reading if the first sentence doesn't do its job. I've been tightening the landing headline around the hotkey/anywhere framing for that reason. Appreciate the offer to send a positioning pass — I'd rather not paste my email into a public thread, so easiest is to reach me via capybro.app, or just reply here and I'll watch the thread. Either way, thank you — more useful than most teardowns.