Connect an AI provider#
Reportworq's AI features do not embed a fixed model. Instead, an administrator configures one or more AI provider connections, each of which names a provider, holds its credentials and model selection, and carries an Enabled flag. Every AI feature, such as insights and the RWAIINSIGHT function, then resolves a model from the set of enabled connections. Configuring at least one enabled connection is the single prerequisite for all AI in the product.
When to use it. Do this once, during setup, before authors add AI narrative to their distribution jobs or use the RWAIINSIGHT function. With no enabled connection, the AI features stay hidden or cannot run.
Before you start#
- Administrator access to Settings, in the AI Connections family of the integrations hub. See The integrations hub.
- Valid provider credentials for the provider you intend to use. The one exception is Reportworq AI, which uses your license entitlement rather than an API key.
- AI connections are global to the instance, only a system administrator configures them. Each connection's Availability decides which workspaces may use it.
Add and enable a connection#
- Go to Settings ▸ Integrations and choose Add integration.
- In the AI Connections group, pick the provider tile. A new connection is created disabled, with provider defaults, and its editor opens.
- At the top of the editor, set the common integration chrome: a Display name (this is how authors pick the connection later, and how the
RWAIINSIGHTfunction resolves it, so make it recognizable), and Availability, the workspaces this connection is visible in. - Fill in the provider's required fields (see the table below).
- Choose Test connection. Reportworq calls the provider and reports success or the provider's own error message.
- Turn on the Enable Integration checkbox at the top of the editor, then Save. Only enabled connections are used by downstream features. Save stays disabled until you have changed something.
Provider reference#
| Provider | Required fields | Notable options and defaults |
|---|---|---|
| OpenAI | API Key, Model Id | Optional Organization Id. Default Temperature 0.7, Default Max Tokens 1024. Model Id defaults to gpt-5.1. Temperature range 0-2. |
| Azure OpenAI | Endpoint, Deployment Name, API Key | Default Temperature 0.7, Default Max Tokens 1024. Temperature range 0-2. Create the Azure OpenAI resource and deploy a model first, then record the Deployment Name, API Key, and Endpoint. |
| Amazon Bedrock | Access Key Id, Secret Access Key, Region, Model Family, Model Id | Optional Session Token. Default Temperature 0.5, Default Max Tokens 1024. Temperature range 0-1. The Model Family dropdown currently offers Anthropic only. Region is the source region for request origination. Supply a Session Token only if your organization's security policy requires one. |
| Google Gemini | Connect with (Gemini API key or Vertex AI service account), then API Key, or Service Account Key, Project Id and Location. Model | Model list loads from Google once credentials are entered. Default Temperature 1.0, Default Max Tokens 2048, Thinking Level low, Thinking Token Allowance 8192. See Connect Google Gemini. |
| Reportworq AI | Model Tier | No API key, access is brokered through your license with managed token quotas. Shows a Token Usage and Quota panel with a Check now action. Available only when your license grants the hosted-AI entitlement. |
Connect Google Gemini#
A Google Gemini connection can reach Gemini in two ways. Pick the one that matches how your organization buys Google AI, using Connect with at the top of the editor.
Gemini API (API key)#
- Open Google AI Studio and sign in.
- Choose Create API key, pick a Google Cloud project (or let AI Studio create one), and copy the key.
- In Reportworq, set Connect with to Gemini API (API key) and paste the key into API Key.
Vertex AI (service account)#
- In the Google Cloud console, select your project and enable the Vertex AI API (APIs and Services, Library).
- Create a service account (IAM and Admin, Service Accounts) and grant it the Vertex AI User role (
roles/aiplatform.user) on that project. - Open the service account, choose Keys, Add key, Create new key, pick JSON, and download the file.
- In Reportworq, set Connect with to Vertex AI (service account). Paste the whole contents of the JSON file into Service Account Key, enter the Project Id (the id, not the display name), and keep Location as
globalunless your organization requires a region such asus-central1oreurope-west4.
Once saved, the key is not shown again. The editor shows the service account's email instead, with a Replace button.
Choose a model#
After you enter credentials, the Model list fills from Google with the text models those credentials can use. It reloads a moment after you change a credential, and Refresh reloads it on demand (the list is otherwise cached for 15 minutes). If the list cannot load, the editor says why, for example that the key was refused or the location was not found. You can always choose Type a model id and enter an id such as gemini-2.5-flash yourself. A saved model that is no longer in the list stays selected, with a warning, so it is never cleared for you. On Vertex AI the list is Google's catalog for the location, so a listed model can still be refused by your project when it runs.
Thinking#
Gemini models think before they answer, and the thinking uses the same output limit as the answer. Default Max Tokens is the size of the visible answer; Reportworq adds the Thinking Token Allowance on top so that thinking cannot use up the answer. Thinking Level (none, low, medium, high) sets how much the model thinks; none sends no thinking setting. Keep Default Temperature at 1.0 unless you have a reason to change it: Google advises that lower values can make newer Gemini models repeat themselves.
Credentials stay with the connection#
A Gemini connection uses only the key or service account saved in its editor. Google environment variables or a gcloud login on the Reportworq server (such as GOOGLE_API_KEY) are never used, so one connection cannot run on another account's credentials. A connection with no key of its own is refused.
Notes and limits#
- OpenAI silently falls back to
gpt-4o-miniwhen Model Id is blank. A new OpenAI connection defaults its Model Id togpt-5.1, but if the stored Model Id is ever cleared to nothing, the connection does not error, it quietly runs ongpt-4o-miniat call time. That can surprise anyone reconciling model choice or billing against what the editor appears to show, so always keep an explicit Model Id set. - Temperature ranges are imposed by the provider, not clamped in the form. The temperature box accepts any number; an out-of-range value is not blocked when you save, but the provider's API rejects it at call time. Stay within the ranges above (OpenAI and Azure OpenAI 0-2, Bedrock 0-1).
- Some providers are preview-gated or off in the current build.
- Amazon Bedrock currently exposes only the Anthropic model family in the editor.
- Azure AI Foundry exists in the code but is not registered, so it does not appear as a choice in the current build. Do not plan around it.
- Credentials are masked. API keys, AWS secret and session keys, and Google service account keys render in a masked field with autofill disabled, and are stored through the AI connection content store.
- Reportworq AI metering. The hosted connection shows monthly quota, used this month, remaining, this session, lifetime total, and the reset date, with a Check now action. Other providers are not metered centrally.
Going deeper. Every AI call made through these connections is recorded. See Audit logs. The internal prompts sent to the provider are editable in The AI prompt store.
Feedback on this page
Comments, questions, requests, or something missing or unclear? Email us - the page you are on is filled in for you.
Email feedback on this pageOr write to support@reportworq.com directly.