Skip to main content

Configure AI Assistant

Connect AI Assistant to a provider approved by your organization.

Free
Unavailable
Pro
Included
Self-hosted
AI tooling bundle
Compare plans

Enable access​

  1. Check that your license includes the AI tooling bundle.
  2. If DISABLED_FEATURES contains ai-assistant, remove it from that list.

Pipeline assistance also requires the Text processing bundle. For license changes, contact support.

Choose who can use it​

By default, AI_ALLOW_ANONYMOUS_USAGE=false requires sign-in and tracks usage per account. Enable accounts and configure sign-in for this mode. If accounts are disabled, users see a message that sign-in is required.

To allow everyone who can access your installation to use the assistant, set:

AI_ALLOW_ANONYMOUS_USAGE=true

This mode disables account usage tracking and account allowances, including for signed-in users. Your provider's charges and limits still apply.

For signed-in usage, configure AI_USER_TOKENS_PER_DAY and AI_USER_TOKENS_PER_MONTH in the supplied template. Set a value to 0 to disable that period's cap. Users see allowance percentages and reset times.

Configure the provider​

Choose an OpenAI-compatible Chat Completions endpoint that supports strict JSON schemas. Set its full endpoint URL, model, and credentials in your deployment .env file:

AI_LLM_ENDPOINT=https://your-provider.example/v1/chat/completions
AI_LLM_MODEL=<model-name>
AI_LLM_API_KEY=<provider-key>

Use the supplied template to set reasoning effort and output limits supported by your model. AI_LLM_TOKEN_PARAMETER selects the provider's output-limit field. If the provider rejects safety_identifier, set AI_LLM_SEND_SAFETY_IDENTIFIER=false.

Contact support for provider compatibility. Keep the provider key in the deployment configuration.

After configuration changes, recreate the services. Then send a request with synthetic examples to check the connection.

Choose where content is processed​

The provider receives the request content described in the user guide. By default, requests also include a hashed account or browser-session identifier for abuse prevention. An external provider receives this content outside your deployment. For an isolated network, choose an internal provider that your deployment can reach. Review your provider's data-processing policy before enabling access.

Help your team get started​

Users open AI Assistant from the right sidebar and send a request. They choose which examples and tests to share, review results, and apply the changes they want. See the user guide for attachments and suggestions.