Guides · Last updated 20 October 2026

Connect your AI provider to Hintglass

To connect Hintglass, a macOS menu-bar app, to your AI provider, pick your host in Settings → Connection, paste your API key, save it to the Keychain and click Test connection.

Hintglass has no AI service of its own: you bring your own key, and your provider bills you for what you use.

Connect

1. Choose Settings… from the Hintglass menu.
2. In Connection, pick your Host. Every preset comes ready: Hintglass already knows its address, how it talks and what it can do.
3. Check the Model. Ollama Cloud and Anthropic start with a model filled in; for the other hosts, type the model’s name as your provider lists it. Hintglass remembers the model for each host, and a session won’t start with it empty.
4. Paste your API key into Key. Ollama (this Mac) takes no key, and Settings says “This host takes no key.”
5. Click Save to Keychain. Key status then reads “Saved — paste a new key to replace it”.
6. Click Test connection. When it passes, the Test row reads “Tested”.

A session starts only on a connection that has passed a Test with exactly these settings. After changing the host, the model, a reasoning level or the key, test again; until you do, a start shows “This connection hasn’t passed a Test with these settings. Run Test connection in Settings → Connection.”

Every host Hintglass supports

Tested on Anthropic (Claude), OpenAI, Gemini, OpenRouter and Ollama Cloud. Built to support them all.

Settings → Connection offers ten hosts and Other. Each one is built for its host: the fields it sends, its reasoning levels and its image limits. Test connection checks them with your key. Here is the interface each one speaks:

Ollama Cloud is the host Hintglass starts with. With Ollama (this Mac), the model runs on your Mac, so its requests stay there.

Use any other host

1. In Host, pick Other.
2. Type the host’s base URL into URL, such as https://example.com/v1. Use https: plain http is only for a server on this Mac. The key goes in the Key field, never in the URL.
3. Pick the Interface the host speaks: Chat Completions (OpenAI-style) or Messages (Anthropic-style).
4. Fill in the Model and the Key, click Save to Keychain, then click Test connection.

Other sends only the fields every host of its interface takes, and Test connection finds the rest.

What Test connection checks

• It sends each kind of request a session sends, with a one-word prompt, to check the key, the model and every field. Each request costs a few tokens.
• It says “Passed.” or “Failed.”, with a line for each step, such as the key, the model and each kind of request, saying what it found.
• It checks whether the model reads images, which ⌥⌘C (Option-Command-C) needs. When it does, Connection says “This model reads images, so code mode is available.”; otherwise it says why code mode is unavailable.
• If the host was slow to start answering, it warns you that “ambient answers will be dropped”: lower the reasoning or choose another model. The warning never fails the Test.

Reasoning levels

Where the host takes a reasoning setting, Connection has one for each kind of answer: Ambient reasoning for the answers you hold the key for, then Code reasoning and Search reasoning. Each starts at a default you can change; more reasoning can delay the first word. A host with no reasoning setting is sent none.

Your key

• Save to Keychain stores the key in your login Keychain. Hintglass sends it only to that host, and the field never shows a stored key.
• To replace it, paste the new key and click Save to Keychain; to delete it, click Remove. Either way, test again.
• Without a saved key, a start says “No API key for OpenAI. Add it in Settings → Connection.”, with your host’s name

What it costs you

• Each answer, each Test request and each profile you generate counts toward your usage with your provider.
• On Ollama Cloud, Ollama (this Mac) and Anthropic, Hintglass also warms the host’s cache in the background, even if you never ask: within a session’s first minute and then once or twice a minute, and while you hold the key, so answers start faster. Your provider may count those requests toward your usage too.

Connection changes apply the next time you start a session; a running session keeps the connection it started with.

Build your Hintglass profile →