RecurseChat – Local AI Chat on Your Mac

TL;DR

TL;DR: RecurseChat is a Mac App Store app that gives you a single, polished interface to run local GGUF models, connect to Claude and ChatGPT, chat with PDFs via local RAG, and summon AI from any app with a global keyboard shortcut — all without a subscription.

Source and Accuracy Notes

⚠️ This section is MANDATORY. All links must be verified from actual source, not guessed.

• Project page: recurse.chat — verified on 2026-06-28
• App Store: RecurseChat on the Mac App Store — verified on 2026-06-28
• HN launch thread: news.ycombinator.com/item?id=39532367 — 637 points, author xyc
• License: Proprietary (one-time purchase, EULA on website)

What Is RecurseChat?

RecurseChat is a macOS application that unifies local and cloud AI chat under one roof. It launched on Hacker News in late 2024 and quickly accumulated 637 points and over 100 App Store reviews. The creator, xyc, describes it as a personal AI app “for using local AI as a daily driver.”

The app is distinct from browser-based chat because it runs as a native Mac app — giving it access to your filesystem, a floating always-on-top window, and native macOS keyboard shortcuts.

Key Features

From the official site:

Local-First Privacy

All model inference happens on your Mac. Conversations with local models never leave the device. Even cloud model traffic goes directly to OpenAI/Anthropic’s APIs — no intermediary servers.

Multiple Model Support

RecurseChat connects to:

• GGUF models (llama.cpp compatible, run locally)
• Hugging Face endpoints
• OpenAI-compatible APIs
• Claude (via Anthropic API)
• ChatGPT (via OpenAI API)

You can import your existing ChatGPT conversation history and continue them with a local model.

Floating Chat

Press ⌘ + ⇧ + Enter to summon a floating chat window from any application. It stays on top of every window — a persistent AI companion as you work.

Local PDF RAG

Upload PDFs, markdown files, or plain text documents. RecurseChat builds a local RAG (Retrieval-Augmented Generation) index using on-device embeddings. Your documents never leave your machine.

One-Time Purchase

No subscription. Pay once, own it forever. This is notable in a market where most AI chat apps demand $15–20/month.

Setup

Download from the Mac App Store

Visit the RecurseChat page on the Mac App Store and purchase (or try free).

Configure a Local Model

1. Open RecurseChat and go to Settings
2. Under Model Provider, choose Local (GGUF)
3. Point to a GGUF model file on your machine — the app works with any llama.cpp-compatible model
4. For Ollama integration, use an OpenAI-compatible endpoint pointing to http://localhost:11434/v1/chat/completions

Connect Cloud Models

In Settings, add your API keys for OpenAI or Anthropic. The app stores keys locally and never transmits them anywhere other than directly to the respective API.

Summon Floating Chat

Press ⌘ + ⇧ + Enter from any app. The floating window appears instantly.

Chat with a PDF

1. Click the + button in the chat bar
2. Select Upload Document
3. Choose a PDF, Markdown, or text file
4. The document is embedded locally — ask questions about its contents

How It Compares

| Feature | RecurseChat | ChatGPT | LM Studio | |---|---|---|---| | Local models | Yes | No | Yes | | Cloud models | Yes | Yes | Yes | | Floating chat | Yes | No | No | | PDF RAG (local) | Yes | Yes (cloud) | No | | One-time purchase | Yes | Subscription | Free | | Mac App Store | Yes | Browser | No |

RecurseChat fills a specific niche: the person who wants a polished, native Mac experience with local AI capability but also wants the flexibility to switch between local and cloud models without juggling multiple apps.

Practical Evaluation Checklist

• [ ] Installs from Mac App Store cleanly
• [ ] Connects to a local GGUF model via Ollama
• [ ] Floating chat shortcut (⌘ + ⇧ + Enter) works from a non-chat app
• [ ] PDF upload and local RAG queries work without internet
• [ ] Cloud model API keys are stored locally, not transmitted elsewhere
• [ ] App works completely offline with local models
• [ ] Import ChatGPT history feature works as described

Security Notes

• Local model inference is fully offline — zero network traffic for local models
• Cloud API calls go directly to OpenAI/Anthropic endpoints
• No data collection or analytics for local conversations (verified from privacy policy)
• Keys stored in macOS Keychain

FAQ

Q: Does it work without internet? A: Yes — local GGUF models run entirely offline. Cloud model features require internet, but the core chat experience does not.

Q: What local models does it support? A: Any GGUF-format model that works with llama.cpp. This includes Llama, Mistral, Phi, Qwen, and thousands of quantised variants from Hugging Face.

Q: Can I import my existing ChatGPT conversations? A: Yes. RecurseChat supports importing ChatGPT export data and continuing those conversations with a local model.

Q: Is there a subscription? A: No. It is a one-time purchase from the Mac App Store.

Q: Does it support Windows or Linux? A: No — RecurseChat is macOS only (Mac App Store distribution).

Conclusion

RecurseChat is one of the cleanest local AI chat experiences available on macOS. The floating chat window alone makes it worth considering for anyone who wants AI assistance without constantly switching context to a browser tab. The one-time purchase model is a refreshing contrast to the subscription-heavy AI market.

If you want to run local models and cloud models from the same app without any ongoing cost, RecurseChat is worth the $15–20 (exact price on the App Store page). The llama.cpp creator @ggerganov called it “great work,” and the App Store reviews are consistently positive about its simplicity and polish.