AI Providers

The chat needs a model, and a provider serves it. inksprite sends your requests straight to the provider you choose; nothing goes through us.

• OpenRouter gives you one account for hundreds of models, paid as you go. Getting started walks through it.
• Anthropic or OpenAI take an API key from your account with them.
• A local model runs on your own computer, free and private, if your computer is up to it.

Each is set up under Settings › AI: choose it under Provider, then choose a Model. The ⋮ beside Provider opens its settings, for a key or another address.

Anthropic and OpenAI

Make a key at Anthropic or at OpenAI. Then in inksprite:

1. Open Settings › AI and choose Anthropic or OpenAI under Provider. It says Not connected: click Connect beside that.
2. Paste the key in API key and click Connect. inksprite says how many models it found.
3. Close the window and choose a Model.

Local models

Provider lists LM Studio, Ollama, llama.cpp, vLLM and KoboldCpp. For any other server that speaks the OpenAI API, see another server.

1. Start your server as its section below says.
2. Open Settings › AI and choose the server under Provider.
3. Choose a Model. inksprite asks the server what it has.

In a browser, some servers have to be told to let inksprite reach them; their sections say how. The desktop app doesn’t need that.

If no models come, or your server is at another address or asks for a key, click the ⋮ beside Provider, change the Endpoint or paste the API key, and click Connect.

LM Studio

Download and load a model, as LM Studio’s guide shows. Then start the server on the Developer tab and turn on Enable CORS in its settings.

• Endpoint: http://localhost:1234/v1
• API key: LM Studio’s token, if you turned on Require Authentication

Ollama

Install Ollama and download a model with ollama pull. To let inksprite reach it, set OLLAMA_ORIGINS to https://app.inksprite.io and restart Ollama, as its FAQ shows for each system.

• Endpoint: http://localhost:11434/v1
• API key: none

llama.cpp

Install llama.cpp and start its server with a model file, or with -hf and a model on Hugging Face. Its guide has the other options.

llama-server -m model.gguf

• Endpoint: http://localhost:9931/v1, or the address the server prints as it starts (older versions use port 8080)
• API key: what you gave --api-key, if anything

To turn a model’s thinking off, set Reasoning effort under Generation to Disabled.

vLLM

Install vLLM and start its server with tool calls on, which the chat needs, and inksprite allowed to reach it:

vllm serve Qwen/Qwen3-8B \
  --enable-auto-tool-choice --tool-call-parser hermes \
  --allowed-origins '["https://app.inksprite.io"]'

The parser depends on the model: vLLM’s list says which to use.

• Endpoint: http://localhost:8000/v1
• API key: what you gave --api-key, if anything

KoboldCpp

Download KoboldCpp and start it with a model file:

koboldcpp --model model.gguf

• Endpoint: http://localhost:5001/v1
• API key: what you gave --password, if anything

Another server

For a server inksprite doesn’t list, click the ⋮ beside Provider, then the + beside the provider list. Give it a Name, fill in the Endpoint and API key, and click Connect. Then close the window and choose a Model.

Presets

A preset is a provider, a model and their settings, kept together. To make another, click the ⋮ beside Preset in Settings › AI and choose New preset. Every chat uses the preset you choose there, or under AI in a chat’s settings (the gear in the message box).

Generation and Sampling show only the settings the chosen model takes.