Start here
AI providers
Use the AI you already have: your claude.ai login, your server, the apps on your computer, free models, or an API key.
The pill at the top of the panel shows who's answering. Click it for the AI picker: providers down the left, their models on the right, and depth chips under models that can think harder. Your pick is used for the next message, and stays until you change it. Settings → AI has a card per provider with its setup and a Test button.
| Provider | Badge | Setup | Sees images |
|---|---|---|---|
| Claude | Your account | Be signed in to claude.ai | Yes |
| Your server | Your server | Address + password | Yes |
| OpenCode | Free | OpenCode app + Bridge, or a Zen key | With a key |
| Claude Code | App on this PC | Bridge | No |
| Cursor | App on this PC | Bridge | No |
| OpenRouter, Gemini, OpenAI, Groq | API key | Paste a key | Most |
| Local model | Your URL | Server URL + model | Model dependent |
#The picker
- Search filters every ready provider's models at once.
- Providers that still need setup are dimmed; picking one shows what's missing and a button to set it up.
- Each model shows its maker's logo, a Free badge when it costs nothing, and how many depths it has.
- Depth chips (Low, Medium, High, Extra high, Max, and Fast where the model has it) set how hard the model thinks. ChroMind remembers the depth per model.
- Lists are cached for six hours; the refresh button next to a provider's name loads them again.
#Claude (claude.ai login)
If you're signed in to claude.ai in the same browser, ChroMind uses that login. Requests go straight from the extension to claude.ai, like the web app, and count against your plan the same way.
- Model: Claude uses claude.ai's default. Paid plans can pick Opus 5.5, Sonnet 5.5, Haiku 5.5 or Fable 5.1.
- History: each ChroMind chat becomes one claude.ai conversation named ChroMind · …, with an Open in claude.ai link under each answer. Turn off Keep ChroMind's chats in your claude.ai history and they're temporary, deleted when you start a new chat.
- Page text is sent as an attachment, so the conversation on claude.ai stays readable.
Signed out, or a browser check?
ChroMind shows Sign in to claude.ai with a button. If claude.ai shows a "checking your browser" page, open claude.ai in a tab once and try again.
#Your server
Hermes Agent, or your dashboard's models (Claude, Cursor and OpenCode on that machine), from the server's address and password. See Your server.
#OpenCode (free models)
OpenCode has free models (Big Pickle and others marked Free). They only answer through the OpenCode app, so ChroMind runs it through the Bridge; no account needed.
- With an OpenCode Zen key in its settings card, every Zen model answers directly, with no app or Bridge, and screenshots work.
- The model list comes from OpenCode, so new free models show up by themselves.
#Claude Code
Uses the Claude Code app with the account it's signed in to
(claude auth login), so your Claude Pro or Max plan answers. Needs the Bridge.
- Opus, Sonnet, Haiku and Fable, each with five depths (Claude Code's
--effort). - All tools are off and it runs in an empty scratch folder with no saved session. It answers; it can't touch your files.
#Cursor
Uses the Cursor CLI (cursor-agent login) with your Cursor plan and its
model list: Auto, plus every model your plan has, grouped with their depths and Fast variants.
- Runs in Cursor's read-only ask mode in an empty scratch folder.
- The first answer takes a few seconds while the app starts.
#API keys
Paste a key in the provider's card (Get a key opens the right page). Keys stay in this browser.
| Provider | Good for |
|---|---|
| OpenRouter | Hundreds of models with one key; free ones are marked Free and listed first |
| Gemini | Google AI Studio, with a free tier |
| OpenAI | GPT models |
| Groq | Open models, very fast |
ChroMind lists only chat models (no embedding, image or audio models).
#Local model
Anything with an OpenAI-style /v1/chat/completions:
| Server | URL | Notes |
|---|---|---|
| Ollama | http://127.0.0.1:11434 | Model like llama3.2; vision models take screenshots |
| LM Studio | http://127.0.0.1:1234 | Start the local server in LM Studio |
| vLLM, LiteLLM, llama.cpp | Your server | Any OpenAI-style endpoint |
ChroMind adds /v1 when the URL doesn't end in a version. Load asks the server for its models.
#Switching mid-chat
You can change provider between messages. Most providers get the whole chat each time. For Claude, ChroMind starts a claude.ai conversation and brings the earlier messages along. Each answer is labeled with the model and provider that wrote it.