Models

Bots think with models, and models come from providers. Connect as many as you like, and give each bot the one that suits its job: a clever model for research, a quick cheap one for sorting the inbox, a local one for private notes.

Admins manage providers in Settings → Models.

#Providers Sugabots knows

Sugabots has presets for these. Pick one and it fills in the address for you.

Where it runs

AllHosted servicesOn your own machine

AnthropicClaude models. Keys are issued at console.anthropic.com.

OpenAIGPT models. Keys are issued at platform.openai.com.

ChatGPTGPT models on your ChatGPT Plus or Pro plan, signed in the way Codex CLI does. Unofficial.

OpenRouterMany providers' models behind one key, issued at openrouter.ai.

Google GeminiGemini models over Google's OpenAI-compatible endpoint. Keys from AI Studio.

GroqFast open-weight models. Keys are issued at console.groq.com.

xAIGrok models. Keys are issued at console.x.ai.

MistralMistral models. Keys are issued at console.mistral.ai.

DeepSeekDeepSeek models. Keys are issued at platform.deepseek.com.

Together AIHosted open-weight models. Keys are issued at api.together.ai.

Fireworks AIHosted open-weight models. Keys are issued at fireworks.ai.

CerebrasFast open-weight models. Keys are issued at cloud.cerebras.ai.

OllamaModels you run yourself with Ollama. No key; the stock install listens on port 11434.

LM StudioLM Studio's local server. No key; it listens on port 1234 once started.

llama.cppA llama-server you run yourself. No key unless you started it with one.

vLLMA vLLM server you run yourself. No key unless you started it with one.

Every workspace starts with Anthropic, OpenAI and Ollama listed, plus any provider you picked during onboarding. Paste a key to use a hosted provider. Local servers need an address, not a key.

#Choose which models bots can use

Each provider lists its models. Switch on the ones you want, and only those show up when someone picks a bot's model. Use Refresh list to fetch new ones, or Add model to type in a model ID.

Each model also says what it can do, like tools, vision or reasoning. Bots use tools, so pick models that support them.

#Behind-the-scenes model

Sugabots also uses a model for its own work: summarising and titling conversations, and deciding which bot replies when nobody was addressed. It starts with none. An admin picks one under Default → System agents use. A fast, cheap model is best. Until then, chats have no summaries or titles.

#Models on your own machine

Ollama, LM Studio, llama.cpp and vLLM run on your own hardware, so your messages never leave it.