Waldo doesn't come with an AI built in — you connect the one you want, and that's a feature, not a hassle. If your experience with AI so far is a chat window in a browser, this page is for you.
When you use ChatGPT or Claude on the web, you're using a chat app: you log in, you type, the AI answers. Behind that app sits the actual AI model — and the companies that run those models also offer a second door to them, built for software: the API (don't worry about the acronym; think "direct line").
An API key is a long password-like code that opens that direct line. You create it once on the AI company's website, paste it into Waldo's settings, and from then on Waldo can talk to the same class of models your chat app uses — right from inside ArcGIS Pro, about your project.
Why does Waldo work this way? Three reasons: you're never locked to one AI vendor (pick Claude, OpenAI, or others, and switch anytime); we never mark up or meter your usage (you pay the AI company directly, at their prices); and your data goes only where you decide — straight from your machine to your chosen provider, with no Waldo server in the middle.
All three take under fifteen minutes. Waldo's settings has a field for each — paste, pick a model from the list, done.
A frontier model with strong reasoning — a great default for GIS work.
Set a monthly spending limit in the console if you want a hard ceiling. You can revoke the key there anytime.
Same idea, different vendor — GPT-class models.
This same settings slot accepts any OpenAI-compatible service — aggregators like OpenRouter (one key, many vendors' models) work by pasting their web address as the endpoint.
The AI runs on your own computer. Nothing leaves your machine, ever.
Honest caveat: local models trail the frontier models in capability, and bigger ones want a modern GPU. But for shops that can't send data to any cloud, this is the option that makes Waldo possible at all — and it costs nothing per use.
Our steps above are the short version. These are the providers' own guides — authoritative and kept current by them.
With an API key you pay the AI company per use — metered like a utility, not a subscription. Prices vary by model and session length, but for perspective: a typical Waldo working session costs cents to a few tens of cents with a mainstream model. A $10 credit lasts most analysts a long time.
Bring-your-own-model is exactly the shape IT departments want AI adoption to take: the AI relationship, billing, and data-governance terms are yours, not a plugin vendor's.
If your organization already has an Anthropic or OpenAI account — or an enterprise agreement with negotiated data-handling terms — IT can issue each analyst an API key under it. Waldo inherits whatever terms your organization already negotiated, because the connection is yours.
Many orgs route AI traffic through an internal OpenAI-compatible gateway or proxy for logging, budgets, and model control. Waldo's endpoint field accepts any such address — analysts paste the internal URL and IT keeps central control of models and spend.
Government, utility, and defense GIS shops that can't use cloud AI can run models on their own hardware — a workstation GPU or an internal inference server — and Waldo works the same way, with no data leaving the network.
Waldo stores keys locally, encrypted with Windows' built-in protection (DPAPI), on each analyst's machine. Keys are never transmitted to Waldo GIS — revoke or rotate them from your provider's console at any time.
Claude subscriptions: no. As of early 2026, Anthropic's consumer terms restrict Pro/Max sign-in to Anthropic's own apps (claude.ai and Claude Code) — third-party software like Waldo is required to use API keys. That's Anthropic's policy, not ours, and it applies to every third-party tool equally. The consolation: an API key is pay-as-you-go and cheap for Waldo-sized sessions, with no second subscription to carry.
ChatGPT subscriptions: a qualified maybe. OpenAI has been notably friendlier to third-party tools working with ChatGPT plans, but hasn't put formal terms around it. Waldo includes a ChatGPT sign-in option that we plan to offer as a beta feature; where providers' policies settle, we'll follow them. If in doubt, an API key is the fully supported path for both vendors.
Waldo ships with recommended defaults and shows your account's available models in a dropdown — you don't need to research model names. As a rule of thumb: pick a current mainstream model from your provider rather than the cheapest one; GIS work rewards the stronger reasoning.
The key is stored on your machine, encrypted with Windows' built-in data protection, and used only to talk to your provider. It never touches a Waldo server. If a key is ever exposed, you can revoke it in your provider's console in seconds — that's a real advantage keys have over passwords.
That's usually a data-governance concern, and it's exactly what the local-model option answers: the model runs inside your network and nothing leaves. Show IT this page and the privacy policy — the architecture (no vendor cloud, your keys, your endpoints) tends to be an easier conversation than a typical SaaS AI tool.
Tell us your situation — solo analyst, university lab, government shop — and we'll point you at the simplest setup.