Any of these will get you chatting. Pick one to start — you can always add more later.

1. a.

   Open The Foundry (click the Foundry icon in the sidebar footer).

2. b.

   Expand the API Keys section and click Add API Key.

3. c.

   Select your Provider, give the key a label, and paste the API key from your provider's dashboard.

4. d.

   Save, then click Test Key to verify it works.

• • NanoGPT has a key type of its own. A NanoGPT connection, image, or embedding profile expects a NANOGPT credential rather than borrowing an OpenAI-compatible one.
• • OpenAI-Compatible can hold a key too. The field is optional and unstarred, so a hosted endpoint that requires authentication works without your having to dress it up as some other provider.

1. a.

   Still in The Foundry, find the Connection Profiles section and click Add Connection Profile.

2. b.

   Select your Provider, choose the API key you just added, and pick a Model (use Fetch Models if the list is empty).

3. c.

   Give it a name (e.g., "Claude Sonnet" or "GPT-4o"), Save, and Test Connection.

A word on what the settings on that form actually do, since a profile is the place where they take effect:

• • Max Tokens and Top P are sent to the provider on every turn, alongside Temperature.
• • Max Context drives the context budget. If you're running a model whose name Quilltap doesn't recognise — a custom Ollama tag, an endpoint of your own — this is the field that tells it how much room there really is.
• • Clearing a numeric option leaves it clear. An empty box shows the default behind it in grey; blank means "let the provider decide," and stays blank.
• • Provider changes don't drag credentials along. An API key doesn't follow a profile onto a different provider, and a base URL doesn't follow onto a provider that shows no such field.
• • Profiles can be tagged, which is a considerable comfort once you have a dozen of them.
• • The Fallback section names an understudy. Once you have a second profile, name it here and a call that fails outright — a rejected key, a rate limit, a provider gone dark — is passed to it instead of ending the turn. It applies to that call only; your next message tries your first choice again. The checkbox below it, which lets Quilltap draft a stand-in of its own choosing, starts off. There is more on all of this in the LLM Connections guide.

Embeddings power semantic memory search — finding memories by meaning rather than exact keywords. Quilltap ships with a built-in TF-IDF system that works out of the box with zero configuration. Check The Foundry → The Commonplace Book — if you see a "Built-in TF-IDF" profile marked as default, you're already covered.

For higher-quality semantic search, you can upgrade to an external provider:

• • OpenAI embeddings — Add a profile in The Commonplace Book using your OpenAI key and the text-embedding-3-small model. High quality, very affordable.
• • NanoGPT embeddings — Add a profile using your NanoGPT key; the available models are discovered live from the gateway, so the dropdown is whatever it actually offers today. The same key already powering your chats does this job too.
• • Other providers — Some OpenRouter and Google Gemini models support embeddings. Check the LLM Connections guide for details.

Quilltap runs background tasks — extracting memories, titling chats, expanding image prompts, and more. These don't need your best model, and using a cheaper one saves real money.

1. a.

   Create a second, cheaper connection profile in The Foundry:

   • • OpenAI: gpt-4o-mini
   • • Anthropic: claude-3-5-haiku
   • • OpenRouter: sort by price for low-cost options
2. b.

   Go to The Foundry → The Salon, enable Cheap LLM, and select your cheaper profile.

Build a character from scratch using Quilltap's character editor, or import a SillyTavern-compatible JSON/PNG card file. Characters define personality, backstory, and conversation behavior — the soul your chosen model will inhabit.

Pick a character, select your cloud connection profile, and you're in. The world's most capable models are now at your service — ready to play any role you've written for them.