01
Works where you write
Use Qelvora from browsers, notes, editors, chat apps, and other macOS surfaces.
Local-first correction for macOS
Polish text without sending it away.
Qelvora corrects, rewrites, and translates selected text across macOS using your local Ollama models. It is fast, native, and private by default.
Mail draft
Hey Laura, I pushed the update. this sentense need a cleaner correction before we send it.
Selected text
Correction ready
Local result - English detected
86% right
7 words
1 correction
This sentence needs a cleaner correction.
this
sentense
needs a cleaner correction
Open-model friendly
Qwen3.6
Gemma
DeepSeek-V4
Kimi K2
Z.aiGLM-5.2
MiniMax-M2.7
TencentHunyuan-A13B
ERNIE 4.5
01.AIYi-1.5
SenseTimeSenseNova
Shanghai AI LabInternLM3
BaichuanBaichuan2
OrionStarOrion-14B
Trusted local stack
Built on tools developers already trust.
Why Qelvora
A correction layer designed for actual Mac workflows.
02
Local by design
Your selected text is corrected by Ollama on your Mac, not by a hosted writing service.
03
Built for difficult apps
Services, clipboard, Accessibility, and OCR fallback help capture text reliably.
Recommended for
A serious writing utility for sensitive workflows.
Clean customer replies, investor updates, and internal notes without moving drafts into a hosted editor.
Keep GitHub issues, support answers, and release notes polished while staying inside your local tools.
Use local Ollama models when a cloud correction service is not appropriate for the text in front of you.
Workflow
Select. Correct. Use.
Select text
Highlight the text you want to clean up in any macOS app.
Trigger Qelvora
Use the menu bar, hotkey, or Services entry where available.
Run locally
The selected Ollama model corrects, rewrites, or translates the text.
Copy result
Review the output and place the clean version back into your workflow.
Local-first
For drafts, customer replies, and notes that should stay yours.
Qelvora is intentionally built around local model inference. macOS permissions are used to capture text and support OCR fallback, not to move your writing into a remote correction pipeline.
Local model runtime
Corrections and translations run through Ollama on your machine.
Transparent project
The Swift source is public and MIT-licensed.
Flexible models
Use recommended models, installed tags, or custom model names.
Guides
Learn local AI writing workflows.
Privacy
Private AI writing assistant for macOS
Why local correction matters for drafts, customer replies, support notes, and sensitive writing.
Ollama Local Ollama text correction on MacHow selected text can move from any macOS app to a local model and back into your workflow.
Models Local LLMs for writing correctionHow to think about Gemma, Qwen, DeepSeek, Kimi, Mistral, Llama, and Phi for local text cleanup.
FAQ
Straight answers before you install.
What is Qelvora?
Qelvora is a private local AI writing assistant for macOS. It corrects, rewrites, and translates selected text using Ollama models running on your Mac.
Does Qelvora send text to the cloud?
Qelvora is built around local model inference. The product is designed for workflows where drafts, customer replies, or internal notes should stay on the machine.
Which models work with Qelvora?
Qelvora uses installed Ollama model tags. Local writing workflows can use model families such as Gemma, Qwen, DeepSeek, Kimi, Mistral, Llama, and Phi.
Who is Qelvora for?
Qelvora is useful for founders, developers, support teams, and privacy-first operators who need fast writing cleanup inside normal macOS apps.
Current release
Install the official macOS build.
The source is free to build. The official DMG is the clean install path with signing, notarization, packaging, and Sparkle updates.
Qelvora needs Ollama running locally. Install Ollama first, launch it once, then download the recommended Gemma 4 4B model from the app.
- Gatekeeper: accepted as Notarized Developer ID
- Runtime: Ollama for macOS
- Recommended model: Gemma 4 4B
- Developer ID: Cyril Dieumegard (923MBLC4X4)
- SHA-256: 318e285f945d096a2da56fd1ae4f8173e6a450fa8082c82e51025ba3bfdeb1fb