Web vs Mac.
One product, two shells. The difference is where your data lives and how you pay for inference.
The fork.
In your browser.
Hosted infrastructure.
Runs on dequid's servers. No install. Messages pass through our infrastructure to reach the LLM provider. Your data lives in a cloud database keyed to your account. Free daily credits included — no API key needed to start.
On your machine.
Direct to provider.
Runs locally. Messages go from your machine directly to the LLM provider — dequid never touches your content. Your personas, vault, and memories live in a local SQLite database at ~/.claudechat/. You bring your own API key.
Your message on the web.
When you send a message on the web app, the request travels:
On web, dequid assembles the system prompt (persona context, memory, vault docs) on our servers and forwards the complete request to the provider. We don't read or log message content for any purpose other than delivering your response and storing your conversation history.
Your conversation history, personas, vault docs, and memories are stored in a cloud database (D1/R2) keyed to your account. They're accessible from any browser, any device.
Your message on Mac.
The Mac app runs a local Node server. When you send a message:
On Mac, dequid never touches your content. The system prompt is assembled locally; the request goes directly from your machine to the provider. Your conversation history, personas, vault, and memories live in a SQLite database at ~/.claudechat/data.db on your machine — not in any cloud.
API keys are stored in the macOS Keychain. The local server reads them at call time; the client never exposes them.
Hosted credits vs. your own key.
The two shells use different inference-payment models. You don't pick a model — it's a property of the shell.
- Web app — hosted credits. dequid holds the upstream provider key. You pay dequid in credits (free daily allowance + Plus/Pro plans + credit packs). No provider account required to start. The model catalog is whatever dequid has wired up; per-model rates are published in-app.
- Mac app — bring your own key (BYOK). You supply API keys for the providers you want (Anthropic, OpenAI, Gemini, HuggingFace, Ollama for local models, and more). Keys live in your macOS Keychain. The provider bills you directly at their rates for inference.
What stays the same across both: personas, SEEK, vault, memory, and scope enforcement all work identically. The product is the same; only the data path and the payment model differ.
Ollama and other local models are Mac-only — web can't reach localhost.
Feature by feature.
| Feature | Web App | Mac App |
|---|---|---|
| No install required | ✓ | — |
| Personas · SEEK · Vault · Memory | ✓ | ✓ |
| Per-persona scope (vault, people, memory) | ✓ | ✓ |
| Full-text search | ✓ | ✓ |
| Your own API key (BYOK) | — | ✓ |
| Works offline / on local network | — | ✓ |
| Data stored on your machine | — | ✓ |
| macOS Keychain for secrets | — | ✓ |
| Hosted credits (no provider account) | ✓ | — |
| Ollama (local models) | — | ✓ |
How to decide.
Pick web if you want to start in under 60 seconds with no install and no API key, or if you work across multiple devices.
Pick Mac if your content is sensitive, if you want your data on your machine, or if you have existing LLM provider accounts and want to use them directly via BYOK.
Back to all playbooks
Five guides, reference, and FAQ.