Connect a model

Kallamo does not come with a model of its own. You bring your own key, or a model running on your computer. There are three stops:

Stop	Where	What it decides
Connections	Settings	Where requests go, and which models you can pick.
AI roles	Settings	The model Kallamo’s own helpers use.
AI profiles	Library	The model each voice replies with.

The setup guide on the launch screen walks through the same steps inside the app, and you can close it and come back at any step.

1. Open Settings

   Settings lives at the top right of the window, on every screen. Its search box finds any option by name.

2. Add a connection

   In Connections, choose Add connection and pick a provider. Then fill in the details, add the models you want, and press Create connection.

3. Choose models for the AI roles

   In AI roles, set the System AI: the default model for Kallamo’s helpers.

4. Give each profile its model

   In the Library, select a profile marked Needs a model, press Set up, pick the connection and a model, and press Save profile.

Providers

Provider	What you need	Notes
OpenRouter	An API key	Many models from many companies with one key. The easiest place to start.
Anthropic	An API key	Claude models.
OpenAI	An API key	GPT models.
Google AI	An API key	Gemini models. Leave the base URL empty for Google’s own endpoint.
Vertex AI	Project ID, region and a service account key	For a Google Cloud project. Paste the whole JSON key file you downloaded.
AWS Bedrock	Access key ID, secret access key and region	For an AWS account, in the region where your Bedrock models are enabled.
Local or compatible	A base URL	LM Studio, Ollama or any server that speaks the OpenAI format. Most need no key.

Name is only a label: it is how the connection appears when you pick a model for a profile or a role. Base URL is where requests go. Leave it as it is for the big providers; change it for a local server or a proxy.

Models: type each model’s ID exactly as the provider writes it and press Add. Profiles and roles choose from this list. You can add more models later from the connection’s menu, and duplicate a connection to try a variation.

Adding a local server?

Choose Local or compatible and set the base URL to your server, ending in /v1: http://127.0.0.1:1234/v1 for LM Studio. Load the model in the server first, then add its ID. Local models are private and free to run, but they need a capable machine, and small models follow long instructions less reliably.

Your keys are encrypted at rest on this computer and sent only to their provider.

AI roles

Kallamo’s helpers work behind the scenes and never write your replies. Each one can use the System AI, its own model, or be turned off.

Role	What it does	Options
Narrative State	Keeps roleplay straight: scenes, what each character knows, relationships. It never writes the story.
Tagger	Marks which Worldbuild entities each passage mentions. It never decides what is canon.	Propose unknown entities: suggests new Worldbuild entries for names it cannot match, for your review.
Retrieval planner	Chooses what to look up before a profile with smart search replies, and stops when it has enough.	Plan every message, Native tool calling.
Summarizer	Titles and summarises archived conversations. The original messages stay searchable.	Archive summaries.

Language for AI notes sets the language of what these helpers write for you to read, such as summaries and analyses.

A fast, inexpensive model is usually enough for the roles. Keep the strong model for the profiles that write: the narrator, the editor. Leaving every role on System AI is a good start.

Costs

Requests go straight from your computer to the provider, billed to your own account. Kallamo keeps the stable part of each request identical between turns, so providers that cache prompts bill less for it. See Context budget and prompt cache.

Next: the workspace.