AI Models
An AI model in MonoChat is a saved connection to one model of one provider: the provider, the model name, the credentials and the default settings the model runs with. Every AI feature picks its model from this list: the Create AI Chat Completion node, the AI Agent node, AI Scopes (for embeddings and voice transcription) and voice-message transcription.
You manage them in Settings › Automation › AI Models.
Models MonoChat provides and your own models
The list shows two kinds of models:
| Kind | Who creates it | Can you edit or delete it? | Who bills the usage |
|---|---|---|---|
| Models provided by MonoChat | MonoChat | No, they are read-only for your workspace | MonoChat, from your workspace balance |
| Your own models | You, with your own API key or endpoint | Yes | Your provider, directly on your provider account |
Models provided by MonoChat are offered to every workspace, so you can start using AI without an account at any AI provider. Their prices are set and managed by MonoChat.
When a flow uses a model provided by MonoChat, the usage (input and output tokens, cached tokens, transcription seconds, embeddings, generated images, audio or video, tool usage) is charged to your balance and appears in Transactions with categories such as AI Input Token, AI Output Token or AI Transcript. A model you create with your own credentials has no MonoChat price: your provider bills you for it.
Supported providers
| Provider | What the form asks for |
|---|---|
| OpenAI | API Key |
| Anthropic | API Key |
| Gemini | API Key |
| OpenRouter | API Key |
| XAI | API Key |
| DeepSeek | API Key |
| Groq | API Key |
| Cohere | API Key, Endpoint (optional, https://api.cohere.ai) |
| Azure | Main Domain (openai.azure.com), API Key, API Version (for example 2024-02-15-preview), Instance Name |
| Ollama | Endpoint (for example http://localhost:11434); the server must be reachable from MonoChat |
| AWS | The provider is listed, but the form currently asks for no credentials for it |
With OpenRouter one API key reaches the models of many vendors. Enter OpenRouter's model ID in Model Name (for example openai/gpt-4o-mini).
Add a model
- Go to Settings › Automation › AI Models and click + (Create).
- Enter a Name and, optionally, a Description. The name is what you will pick in nodes and scopes.
- Choose the Provider.
- Under AI General Settings, select the Features this model supports (see below), and for text models the Response Formats it can return.
- Under AI Model Client Config, enter the Model Name exactly as the provider names it, then the provider's credentials.
- Adjust the generation settings if you need to, and click Save.
The model now appears in every model selector whose feature it has.
When you edit a model, enter the API Key again. MonoChat never sends a saved key back to your browser, so the field is empty on the update form.
Fields
AI General Settings
| Field | Description |
|---|---|
| Features | What the model can do: Text, Embed, Transcript, Tools (tool calling), Vision (reading images), Generate Image, Generate Video, Generate Audio and Reasoning (see below). Selectors that need a feature offer only models that have it: AI scopes ask for Embed, the Speech to Text AI Models selectors for Transcript, the AI Agent node for Text. MonoChat also refuses a call the model is not marked for, for example AI tools on a model without Tools. |
| Response Formats | Shown for text models. The structured outputs the model supports: text, json-schema (a JSON Schema you define in the AI node) and message-schema (a MonoChat message payload, such as buttons or lists). |
AI Model Client Config
| Field | Shown for | Description |
|---|---|---|
| Model Name | All | The provider's model identifier, for example gpt-4o-mini. |
| Credentials | Depends on the provider | See Supported providers. |
| Max Tokens | Models without Reasoning | Upper limit of tokens in a reply. |
| Max Completion Tokens | Reasoning models | Upper limit of tokens the model may generate, including its reasoning. |
| Reasoning Effort | Reasoning models | none, minimal, low, medium, high or xhigh. Higher effort gives better answers to hard questions, at the cost of time and tokens. Leave it empty to use the provider's default. |
| Verbosity | Reasoning models | low, medium or high: how long the answers should be. |
| Cache | All | On by default. Lets the provider reuse cached prompt parts, which lowers cost and latency for repeated prompts. |
| Max Retries | All | How many times a failed call is retried (1–10). |
| Temperature | Models without Reasoning | 0–1. Lower is more predictable, higher is more creative. |
| Top P | Models without Reasoning | 0–1. Limits sampling to the most likely tokens. |
The Create AI Chat Completion node has its own Max Tokens, Max Completion Tokens, Reasoning Effort, Verbosity, Cache, Max Retries, Temperature and Top P fields. Values you set on a node override the model's defaults for that node only.
Where models are used
| Place | Which models it offers |
|---|---|
| Create AI Chat Completion node | Models for the reply, and a Speech to Text AI Models selector for voice messages |
| AI Agent node | Main, fallback and fast models, and a speech-to-text model. See AI Agents. |
| AI Scopes | A model with Embed to turn documents into vectors, and a speech-to-text model for audio files |
Permissions
Access is controlled by the role permissions Read / Create / Update / Delete AI model clients in automation. See Roles.