Home / Features / Your AI, your key

Your AI, your key — or your own server

Bring your own AI — Claude, ChatGPT, or a model you host

Unlock the AI features with Pro, then run them on a back-end you bring: a Claude or ChatGPT key, or your own self-hosted model. Written brand descriptions, premium design manuals, palettes from a brief, and color naming — your account, your data, going straight from your Mac to your provider or your own machine.

Claude (Anthropic)sk-ant-…7f2qIn use
ChatGPT (OpenAI)sk-…9k1dStored

How it works

  • Three back-ends, one active at a time — a Claude (Anthropic) key, a ChatGPT (OpenAI) key, or a custom OpenAI-compatible endpoint. Configure any of them and switch one on in Settings.
  • Run your own model — point Pocket CVI at Ollama, LM Studio, vLLM, or any OpenAI-compatible server. Give it the base URL and model name; the key is optional, since most local servers don't need one. Nothing you don't pay for and nothing that leaves your machine.
  • Keys live in your Keychain — device-local, never synced, never sent to us. Requests go straight from your Mac to Anthropic, OpenAI, or your own server.
  • Pause without losing anything — switch AI off entirely and everything stays saved; the app falls back to its local behavior everywhere.
  • Pick your quality — Best Quality writes with Claude Opus or GPT-5; Fast & Cheap uses Claude Haiku or GPT-5 mini. (A self-hosted model uses whatever you loaded.)

One caveat: official-guide web search needs Claude or ChatGPT — a local model can't browse the web, so that single step declines rather than guess. Everything else works on any back-end.

What a key unlocks

Five things, each marked with a ✨ sparkle wherever it appears:

  • Brand descriptions — written into a website import's CVI notes.
  • Official brand guides — found on the web and filed into the kit.
  • Premium design manuals — written and typeset in the brand's own fonts.
  • Palettes from a written brief — describe the brand, get a named five-role scheme.
  • Color naming — evocative, brand-aware names for unnamed colors, even automatically as your clipboard captures them.

Honest by design

Every surface tells you its AI state before anything runs: an accented sparkle when AI is on (“using your API key”), a gray one when a key would make it better. Verification checks not just that your key is valid but that your account has credit — and if it runs out mid-week, the app says so in your provider's own words, with a link to their billing page.

Attribution is honest too: a design manual's footer says “Generated with Pocket CVI using Claude AI” only when Claude's prose is actually in it.

What it costs

Two things, kept deliberately separate and honest:

  • Pro unlocks the AI features in the app — €5 a month, or €99 once for a lifetime license. Everything else in Pocket CVI is free forever.
  • You bring the model. With a Claude or ChatGPT key your provider bills the usage — typically a few cents per request (a brand description or named palette is a fraction of a cent; a premium manual a few cents). Or run a model on your own hardware with Ollama or LM Studio and pay nothing per request.

There is no markup, no middleman, and no proxy — which is exactly why Pro can be this inexpensive. Bring a developer API key from console.anthropic.com or platform.openai.com, or just point Pocket CVI at your own server.

FAQ

Is a Claude Pro or ChatGPT Plus subscription enough?

No — chat subscriptions don't include API access. You need a developer API key (console.anthropic.com for Claude, platform.openai.com for ChatGPT) with its own pay-per-use billing. Settings links you to the right pages.

What happens without Pro or without a key?

The free app is fully usable: picking, brand kits, website import, the local palette generator, contrast, sharing, and sync. The AI surfaces show a grey sparkle marking what Pro and a key would add — nothing else is held back.

Does Pocket CVI see my key or my data?

Never. Keys are stored in your Mac's Keychain and requests go directly from your Mac to Anthropic or OpenAI. Nothing is proxied through our servers.

Can I switch between Claude and ChatGPT?

Yes — keep both keys stored and flip the Use switch in Settings. Every feature names the agent doing the work, so a manual written by ChatGPT says so.

Can I use my own local model, like Ollama?

Yes. Under Settings → AI, add a Custom / Local endpoint: the base URL (e.g. http://localhost:11434/v1 for Ollama), the model name (e.g. llama3.1), and an optional key most local servers don't need. It works with Ollama, LM Studio, vLLM, LocalAI, or any OpenAI-compatible server — so you can run the AI features with no per-request cost and nothing leaving your machine. The one exception is official-guide web search, which needs Claude or ChatGPT, since a local model can't browse the web.

Why do I pay for Pro and for my own API key?

Because we never touch your data. Pro pays for the app and its AI features; your provider bills the model usage directly. Bundling the AI would mean proxying your brand material through our servers and charging a markup — this way your prompts stay between your Mac and Anthropic or OpenAI, and Pro stays €5.

Related guides

Try it yourself

Download Pocket CVI