Correctify uses YOUR own API key: you pick your AI provider, and nothing goes through a middle-man server. Here is, step by step, how to create a key with each supported provider.
An API key is a personal code that lets the app use the AI on your behalf. You create one only once, paste it into Correctify’s options, and you’re done. Pick whichever provider suits you: Google Gemini has a free tier, perfect to get started.
You don’t have to select anything
Correctify recognizes the provider automatically from the start of the key: a key beginning with “AIza…” is treated as Google Gemini, “sk-ant-…” as Anthropic (Claude), and “sk-…” as OpenAI (ChatGPT). Paste, save, done. In the Windows and macOS app, two drop-down lists also let you pick the provider and the model yourself, should you want to.
Choosing your model
A provider offers several models: a small one, fast and very cheap, and larger ones, sharper but pricier. Correctify picks the most affordable model of your provider by default, which suits text correction very well. To change it, the “Model” list in the options shows what is available. That list is kept up to date from correctify.fr: when a provider retires a model, it disappears from the list and you fall back to the default model automatically, with nothing to do.
Google Gemini
Free tier available
The easiest way to start: Google AI Studio offers a free tier that’s more than enough for everyday text correction.
Cost per correction : Free within Google AI Studio’s free tier, which is enough for everyday use. Beyond that, with the Flash-Lite model selected by default: about $0.0002 per correction, i.e. ~$0.23 per 1,000 corrections.
1
Open Google AI Studio and sign in with your Google account.
2
Click “Get API key” in the top-left.
3
Click “Create API key”, then agree to create or select a Google Cloud project if prompted.
4
Your key appears (it starts with “AIza…”). Click the copy icon.
5
Paste it into Correctify’s options, in the “API key” field. Done.
To use Anthropic’s Claude models. Like OpenAI, the API is billed per use: add a little credit to your account.
Cost per correction : With the Haiku model, Anthropic’s cheapest and the one selected by default: about $0.002 per correction, i.e. ~$2.10 per 1,000 corrections.
1
Create an account (or sign in) on the Anthropic console.
2
In “Billing” (Plans), add a payment method and some credit.
3
Open “API Keys”, then click “Create Key”.
4
Name the key, confirm, then copy it (it starts with “sk-ant-…”): keep it safe, it’s shown only once.
5
Paste it into Correctify’s options, in the “API key” field.
Credit / payment card required Windows and macOS app
The French provider, for those who would rather use a European model. Its keys have no recognizable prefix, so pick “Mistral AI” in the “Provider” list in the options.
Cost per correction : Pay as you go. The rate depends on the model you pick; Mistral’s small models are among the cheapest around. See the pricing page in the Mistral console.
1
Create an account (or sign in) on the Mistral AI console.
2
Enable pay-as-you-go billing and add a payment method.
3
Open “API Keys”, create a key and copy it right away.
4
In Correctify’s options, pick “Mistral AI” in the “Provider” list.
5
Paste the key into the “API key” field, then pick a model in the “Model” list.
Credit / payment card required Windows and macOS app
A middleman giving you access to models from dozens of providers with a single key and a single account. Handy to try several models without opening an account everywhere.
Cost per correction : One key, hundreds of models, each with its own rate shown in the OpenRouter catalogue. A few models there are even free, with limited quotas.
1
Create an OpenRouter account and add some credit (a few euros cover thousands of corrections).
2
Open the “Keys” page, create a key and copy it.
3
Paste it into Correctify’s options: the “sk-or-v1-” prefix is recognized on its own.
4
In the “Model” list, pick the one you want to use.
Estimates for a ~200-word text, using the most affordable model Correctify selects by default at each provider. The math counts both the request (instructions + text ≈ 650 tokens) and the response (corrected text + list of fixes ≈ 400 tokens). If you pick a larger model yourself in the options, the cost rises accordingly: check the provider’s pricing page then. Prices as observed in early 2026, subject to change.
No API key: running the AI on your own computer
The Windows and macOS app can also talk to an AI server running at your place. There is then no key, no account and no bill: your text simply never leaves your machine. It does take a reasonably powerful computer, and quality depends on the model you install.
Ollama
The most common tool to run a model locally. Install it, download a model with “ollama pull”, then pick “Ollama (local)” in Correctify’s options: the app queries your server and offers the list of models you have installed.
An app with a graphical interface, friendlier when you are starting out. Turn on its local server, then select “LM Studio (local)” in Correctify. If you changed the port, just fix the suggested address.
Running an OpenAI-compatible gateway such as LiteLLM or vLLM, or the API of a project of your own? Pick “Custom server”, enter its address, and Correctify will fetch the list of models from it. That address stays on your computer: our site never sees it.
These three options only concern the desktop app. The Chrome, Firefox and Thunderbird extensions work with a Google Gemini, OpenAI or Anthropic key.
What next?
Open Correctify’s options (right-click the icon → Options, or the extension’s gear), paste your key into the “API key” field, and save. The provider is detected on its own; in the desktop app you can then adjust the provider and the model in the two lists provided. Your key stays on your device and is never sent to Correctify.
Keep your key secret
An API key is like a password: don’t share it with anyone and don’t post it anywhere. If you think it leaked, delete it from the provider’s dashboard and create a new one. If you really must pass it to someone, do it through a secure link with seecret.it.
This site uses audience measurement cookies (Google Analytics) to see which pages are useful. The texts you correct are never involved. Learn more