AI Settings

AI Settings are where an administrator picks the AI provider the instance runs on, the models it uses, and the global system instructions every workspace inherits. They live in Admin Console -> Settings -> AI and apply to the whole instance.

Self-hosted instances only

On Mixpost Cloud these settings are managed for you. What your team can do with AI is described under AI Assistant.

Default AI Provider

Once you've connected one or more AI services, you need to set a default provider that will be used across your instance.

1. Navigate to Admin Console -> Settings -> AI
2. Select your preferred AI Provider from the dropdown.
3. Select the Model and, optionally, the Fast model you want to use.
4. Save the changes.

Screenshots:

The selected provider and models are used for all AI Assistant interactions across all workspaces.

Models

Each provider exposes two model tiers so you can balance quality against cost:

• Model — the default, higher-accuracy model. It powers the AI Assistant and any task that needs nuanced output.
• Fast model — a cheaper, faster model used for low-stakes, high-volume tasks.

Leaving a field on Default uses the provider's recommended model for that tier.

Custom Model ID

Both the Model and Fast model dropdowns include a Custom model ID option. Choose it to type any model identifier the selected provider supports — useful for accessing newly released models that aren't in the list yet, or provider-specific deployments.

tip

Enter the exact model identifier expected by the provider's API (for example claude-opus-4-8 or gpt-4.1-mini). An invalid identifier will cause AI requests to fail.

Image Generation

Image generation lets workspace users create images from a prompt in the Media Library. It is off until an image-capable provider is selected here.

Requirements

• An image-capable AI service. Only OpenAI and Gemini support image generation. At least one of them must be connected and active under Admin Console -> Services.

Enable and Configure

1. Navigate to Admin Console -> Settings -> AI.
2. Scroll to the Image section.
3. Choose an Image provider and, optionally, an Image model (see below).
4. Save the changes.

Once saved, an AI button appears in the Media Library's upload area for every workspace.

Image Provider

The Image provider dropdown lists only providers that support image generation (OpenAI and Gemini).

• Select a provider to always use that specific service for image generation, regardless of your default AI provider.
• Leave it empty to fall back to the default AI provider above — but only if that provider supports image generation. If your default provider is text-only (for example Anthropic), image generation stays disabled until you pick an image-capable provider here.

note

The chosen AI provider must also be connected and Active under Admin Console -> Services. The provider selector only controls which service is used — it does not supply the API credentials.

Image Model

The Image model dropdown lists the image models available for the selected provider, plus a Custom model ID option.

• Leave it empty to use the provider's recommended default image model.
• Pick a model from the list to pin a specific one.
• Choose Custom model ID to type any image model identifier the provider supports — useful for newly released models that aren't in the list yet.

Supported Providers and Models

Provider	Image models	Default model
OpenAI	gpt-image-2, gpt-image-1.5, gpt-image-1, gpt-image-1-mini	gpt-image-2
Gemini	gemini-3.1-flash-lite-image, gemini-3.1-flash-image, gemini-3-pro-image, gemini-2.5-flash-image	gemini-3-pro-image

Troubleshooting

The AI button doesn't appear in the Media Library. Image Generation is disabled. Confirm that an image-capable provider is selected in Admin Console -> Settings -> AI -> Image (or that your default AI provider supports images), and that the provider is Active under Services.

Generation spins forever and never completes. The most common causes are a queue worker that isn't processing the mixpost-media queue, or an array cache driver. Verify your worker is running and that CACHE_STORE points to a shared store.

Generation fails immediately. Check that the provider's API key is valid and the account has image-generation access and available credits. Administrators see the underlying provider error message; other users see a generic message.

Engagement (Inbox) Models

The Inbox automatically classifies incoming messages — applying labels, detecting sentiment and spam, and moderating hate speech. You can choose which models power this classification independently of the AI Assistant, so the high-volume classification workload can run on cheaper models without affecting content generation.

These models are configured once, at the instance level, by an administrator. They are not exposed in the per-workspace Inbox settings.

note

The Engagement (Inbox) section appears only when the Inbox Engagement feature is enabled for your instance. If the feature is disabled, these fields are hidden.

1. Navigate to Admin Console -> Settings -> AI
2. Under Engagement (Inbox), select:
   • Engagement model — a higher-accuracy model used for labelling and hate-speech moderation.
   • Engagement fast model — a cheaper model used for sentiment and spam detection.
3. Save the changes.

Leave either field on Same as Model / Same as Fast model to reuse the general models configured above.

Global System Instructions

Global system instructions are instance-level directives that guide the AI's behavior across all workspaces and users. These instructions are applied to every AI interaction, regardless of which workspace is being used.

Use global system instructions to:

• Set content policies or restrictions at the instance level.
• Specify formatting preferences that apply universally.
• Enforce compliance or brand guidelines across all workspaces.

How to Configure

1. Navigate to Admin Console -> Settings -> AI
2. Enter your instructions in the Global System Instructions field.
3. Save the changes.

Screenshots:

tip

Global system instructions are different from Brand Voice. Global instructions apply at the instance level, while Brand Voice is configured per workspace by the people using it, to maintain each brand's unique identity.