Providers & Models

TomoriBot connects to external AI providers rather than hosting a built-in model. You can connect hosted services like Google Gemini, OpenRouter, and NovelAI, or point her at local self-hosted endpoints. You need at least one provider to start chatting.

Subscription

On hosted TomoriBot installations with Subscription enabled, use /subscribe to link your Patreon membership and view your remaining allowance. A member with Manage Server can choose Sponsor this server in its overview or + Sponsor Current Server under Sponsored Servers. /providers also offers sponsorship. If your Patreon account is connected to a Discord account, that account must be the one running /subscribe. Otherwise TomoriBot refuses the link: run the command from the connected account or change the connection on Patreon. The success page names the Discord account it linked. Each server has one active sponsor. Server already sponsored is disabled until that sponsorship is removed. You can sponsor multiple servers; they share your allowance with your personal usage.

API Keys

Add a provider key during first-time setup with /setup, or later from /providers by choosing + Add new Provider. Keys are encrypted at rest, so no one, including server admins, can read them back.

In /setup, choose a hosted provider, a custom endpoint, or User BYOK for a server where members supply their own providers. Setup saves changes when you finish. Use /providers to replace a key; saved sampler settings are preserved.

Each provider has its own key-generation steps. In /help, choose Setup, then Get an API Key, and pick your provider for a step-by-step walkthrough, or use these starting points:

Provider	Notes	Get a key
Google Gemini	Free tier, runs every feature. Recommended first setup.	AI Studio
OpenRouter	Paid models through one key.	OpenRouter keys
NovelAI	Subscription; uncensored storytelling and roleplay (text only).	NovelAI
DeepSeek	Pay-as-you-go reasoning models.	DeepSeek
NVIDIA NIM	Hosted text, embeddings, and image.	NVIDIA Build
Anthropic	Claude models via the API (not Claude Code).	Anthropic
Z.ai	GLM family. 🟡 ToS restricts usage to coding and agent scenarios.	Z.ai
Vertex AI	Google Cloud via gcloud ADC (best for locally-run or dev setups).	see below
Vertex AI Express	Google Cloud API-key BYOK (Preview, Gemini subset).	Express Mode
Custom	Any OpenAI-compatible endpoint (Ollama, vLLM, LiteLLM, …).	see Custom Endpoints

Caution

Never share your API key with anyone else. Add or replace a custom endpoint’s Bearer auth token from its Edit Endpoint action in /providers.

Vertex AI authenticates with Application Default Credentials (ADC) rather than a stored secret. For local hosting, ADC can come from gcloud; hosted deployments should use a workload identity or service account. An AI Studio API key alone does not authenticate full Vertex AI. The selected Google Cloud project must have billing and the Vertex AI API enabled, and the host identity needs Vertex access. Whoever hosts TomoriBot can limit Vertex AI to certain servers and users; everyone else is asked to choose a different provider. The setup guide is available from Google Vertex AI on the API Keys page in /help.

Brave Search is separate from your AI provider. Add its key in /providers to enable more search categories; see Web search.

Choosing Models

Use /providers to manage server credentials, model catalogs, and endpoint registrations. Then use /config > Models > Switch Models to select the shared capability assignments every member of the server uses. Both commands require server management permissions.

Members can set up their own providers and models with /personal providers and /personal config. Personal settings follow them across servers; see Personalization.

New OpenRouter configurations use DeepSeek Flash Latest for chat and Gemini 3.1 Flash Lite for the vision fallback. Both use OpenRouter credits. Replacing a key preserves your vision selection, including - No Model.

In /config > Models > Switch Models, you can assign models and endpoints across these capability slots:

• Text: the main chat model.
• Vision: reads images when the chat model cannot.
• Embeddings: powers the document knowledge base.
• Decision: a registered Decision model. Natural Nudges uses it to decide when to join a conversation.
• Standard Image: standard image generation (see Image Generation).
• NovelAI Image: NovelAI image generation.
• Video: video generation.
• TTS (Voice Design): speaks using a saved voice description.
• TTS (Voice Clone): speaks using a reference sample or provider voice.
• Speech to Text: transcribes audio attachments with a hosted transcription model or a custom endpoint. Pick a provider, then its model; - No Model stops transcription on the server’s key.

Tip

- No Model in the Vision slot does not make a vision-capable chat model blind: it can still read images directly. The Vision slot provides a separate model for image analysis when the chat model cannot see images.

Speech has separate voice design and voice clone slots. See Voice: TTS & STT for setup and provider support. You can also manage backup keys in /providers.

Tip

Adding one backup key to an AI provider turns on rotation for that provider. Requests rotate across the primary key and its backups, and a key that hits a rate limit is skipped for 60 seconds.

Decision Models

Decision models answer structured questions, such as yes/no or a choice from a list. They are separate from the chat model. Natural Nudges can use one to decide when a persona should join a conversation.

Register an OpenRouter Decision model or a compatible custom endpoint in /providers or /personal providers. Select the server’s model under /config > Models > Switch Models. Custom services must support System One or OpenAI Decisions compatibility. Register models from the service’s documentation; a successful connection does not verify their capabilities.

Custom Endpoints

Custom endpoints connect self-hosted or proxy-backed services such as Ollama, LM Studio, vLLM, ComfyUI, and local speech servers. Register a server endpoint in /providers or a personal endpoint in /personal providers.

Choose the API compatibility supported by your service, then register its exact model and supported capabilities. A connection can group models that share one endpoint URL. Personal local speech endpoints are not supported.

For setup instructions, see Local LLM, ComfyUI, and ChatMock.

Supported Providers

If you do not have the hardware to host your own models, TomoriBot supports a wide range of cloud services. Not every feature is available on every provider.

LLM Providers

Provider	Streaming	Tool Calling	Image Input	Embeddings	Notes
Google Gemini	✅	✅	✅	✅	Free models available
OpenRouter	✅	✅	✅	✅	Uses OpenRouter credits
Anthropic (API)	✅	✅	✅	-	Not Claude Code
NovelAI	✅	-	-	-	Text models cannot use tools; see NovelAI
NVIDIA NIM	✅	✅	✅	✅	Free models available
DeepSeek	✅	✅	-	-	-
Z.ai	✅	✅	✅	-	Free models; 🟡 ToS = coding and agent use only
Z.ai Coding	✅	✅	-	-	Subscription plan
Google Vertex AI	✅	✅	✅	✅	Includes ‘free’ Express version
Codex CLI (via ChatMock)	✅	✅	✅	-	Setup

Image Generation

Provider	Text-to-Image	Image-to-Image	Inpainting	Notes
Google	✅	✅	-	-
OpenRouter	✅	✅	-	-
NovelAI	✅	✅	✅	Can combine with other providers
NVIDIA	✅	-	-	Text-to-image only; reference images are ignored
Z.ai	✅	-	-	-

These are default capabilities. When registering an image model, declare the modes its API supports; unsupported modes fail at generation time. NovelAI uses its own image pipeline.

Video Generation

Provider	Text-to-Video	Image-to-Video	Notes
Google	✅	✅	Async polling workflow
OpenRouter	✅	✅	Async polling workflow
Z.ai	✅	✅	Async polling workflow

Voice & Audio

Provider	Text-to-Speech	Speech-to-Text
ElevenLabs	✅	✅
Google Gemini	Voice Design	✅
Vertex AI (standard, global)	Voice Design	✅
OpenRouter Fish	Voice Clone	No
OpenRouter Seed Audio	Voice Design	No
OpenRouter transcription models	No	✅

Local voice engines are covered under Self-Hosting. For built-in web search and URL reading, see Tools & Extensions.