Choose who answers
The built-in assistant, for any game, and the AI help in the binding guide, for DCS aircraft, use a model you choose, with your own key. RigReady has no account or server of its own and never sees your key or your requests. Set it up under Configure > Settings > AI assistance.
The choices
| Who answers | What you need | Where requests go |
|---|---|---|
| Anthropic (Claude), the default | An Anthropic API key | api.anthropic.com |
| OpenAI | An OpenAI API key | api.openai.com |
| OpenAI-compatible, preset Google Gemini | A key from Google AI Studio | generativelanguage.googleapis.com |
| OpenAI-compatible, preset OpenRouter | An OpenRouter key | openrouter.ai |
| OpenAI-compatible, preset Mistral | A Mistral key | api.mistral.ai |
| OpenAI-compatible, preset Ollama | Ollama running on this PC, no key | localhost:11434, this PC only |
| OpenAI-compatible, preset LM Studio | LM Studio's server running on this PC, no key | localhost:1234, this PC only |
| OpenAI-compatible, Other | The service's address and key (Groq, xAI, DeepSeek and others) | The address you type |
Set it up
- Under Who answers, pick the provider.
- For OpenAI-compatible, pick a Service. It fills in the API address and suggests a Model; every field stays editable. For Other, type the address, for example
https://api.groq.com/openai/v1, and press Enter. - Paste the key and press Store key. The page then says the key is stored and shows only its last four characters. A server on this PC needs no key.
- Check the Model. Pick one from the list or type any model id the service offers.
- Press Test. RigReady sends one small request and says plainly whether it worked: "Works. The key works with Claude Opus 5.5.", or why not ("did not accept this key", "does not offer that model").
Each provider keeps its own key, so you can switch between them without pasting again. A key for an OpenAI-compatible service is kept for that service's host only: change the address to another host and the key is removed, so it can never be sent there.
Structured answers
Suggestions and drafted guides come back as JSON, which RigReady checks line by line before you see anything. Most services can be held to a JSON schema. For one that cannot (DeepSeek, most Groq models, some OpenRouter models, older local servers), set Structured answers to Ask for JSON: the schema is then written into the request, the answer is checked the same way, and if it does not check out the same request is sent once more. The preview before each request says when this applies. If the second answer still does not check out, nothing is used and the message says so.
Which host receives what
What is sent depends on what you use, and is the same for every provider.
- AI help in the binding guide is for DCS aircraft only; if you do not fly DCS, none of this is ever sent. It sends the aircraft, its actions and what each is bound to now, the product names and layout of the controllers used in DCS, the guide for the aircraft when there is one (RigReady's, one drafted by AI, or your own) and your question. It never sends file paths, device IDs, serial numbers, the names you gave your devices, your Windows user or PC name, or any other content of your files. See What is sent.
- The built-in assistant works the same whatever you play: a racing sim such as iRacing, a flight sim such as DCS, or anything else. It sends your messages, its own answers, RigReady's instructions and tool list, and what each tool it used answered. That can be anything RigReady itself can show you: your setups, what is connected and running, where programs are installed, the names of your devices, bindings and settings. In what the tools answered, the path of your Windows user folder is written as
{USER}and the PC's name as{PC}; other names and paths go as they are. See The built-in assistant.
Exactly what is sent in Settings says the same, use by use, for the provider you chose.
- Requests go only to the host in the table above (or the one you typed), and only when you press Send on the preview, or Test in Settings. In the assistant panel you can switch the preview off for one conversation; its requests then go as soon as they are ready, until you press Stop.
- The address must start with
https://. Plainhttp://is accepted only for this PC (localhost,127.0.0.1), so the key and your data are never sent unencrypted over a network. - If a service answers with a redirect to another address, RigReady stops and says so instead of following it with your key.
- The key goes only in the request's header. It is encrypted for your Windows account and never written to settings, logs, backups, shared setups or screenshots.
- With Ollama or LM Studio nothing leaves the PC, unless that server is set up to send it on.
- The provider's own terms and privacy policy apply to what you send it.
Tokens and estimates
After each answer, RigReady shows the model, the tokens in (and how many came from the provider's cache) and out, as the provider reports them. Where prices are set for the model, it also shows an estimate in dollars, worked out from those tokens and the prices in Settings, and says which prices those are: RigReady's built-in ones with the month they were recorded, or your own. It is an estimate, not a bill: your provider's console shows what you are charged.
RigReady has built-in prices for the Anthropic and OpenAI models it lists. For other services and models it has none, and shows no dollar figure until you type the model's prices in Settings; check your account with the service. Nobody charges for a model on your own PC, and RigReady shows no dollar figure for it unless you type prices.
The prices can be edited for each model, and a ceiling for one question and a monthly budget can be set, in Settings under AI. See Estimates, prices and limits.