# Shared AI service connections Reuse website connections for external AI prompts, tracing, memory and tools across the features that need them. *Release 3.3 · admin, developer* **Settings › AI › Connections** · **Settings › AI › Features** · **Settings › AI › Functions** ## What it is **Connections** store named website connections to external AI-related services. A connection can supply prompts, tracing, memory, privacy checks, guardrails or tools, depending on its type. Several AI features can refer to the same connection. ## Who uses it and why Administrators choose the external services their AI features should use. Developers provide connection requirements and check that the selected service supports the intended work. Powerfront manages connection creation, credentials, testing and removal for your website. ## Where to find it Powerfront manages this setup in **Settings › AI › Connections**. Ask Powerfront to create or change a connection for you. With the **Settings** role and **AI** permission, or the **Developer** role, you can select available connections where the relevant feature exposes them, such as **Tracing** or **Memory** in **Settings › AI › Features**, and **Tool Connection** in **Settings › AI › Functions**. ## How it works Each connection has a **Name**, a **Type** and provider-specific details. The connection list shows **Capabilities**, such as prompts, tracing, memory and tools. These describe what the connection type supports. | Connection type | Main use | |---|---| | **LangFuse** | Managed prompts, tracing and experiments | | **Mem0**, **INSIDE Memory Service** | Persistent context for AI interactions | | **MCP Server** | Remote tool discovery and execution | | **UCP Connection** | Commerce service discovery and supported operations | | **Presidio (Local)** | Privacy processing | | **Amazon Bedrock Guardrails** | Input and output guardrail checks | | **Activepieces** | Automation flows called by AI functions | | **Salesforce AI Connection**, **Salesforce Agentforce** | Salesforce AI credentials and agent connections | Saving or deleting a connection clears INSIDE's cached clients for that website. A feature or account still needs the appropriate connection selected before it can use the service. ## Configure it Give Powerfront the required service type, a recognisable connection name and the capabilities you need. Supply credentials through the agreed secure process. The details depend on the provider; LangFuse needs its public and secret keys, while memory services need an API credential. Powerfront saves the connection and uses its **test** action to check the supported connection test. Some services return additional information, such as discovered tools. Use the appropriate AI feature or function editor to select the saved connection. Ask Powerfront to update credentials or remove unused connections. Before removal, identify features, functions and accounts that still depend on the connection. ### Fields | Field | Where | Required | What it is for | |---|---|---|---| | Name | Settings › AI › Connections | Yes | Identifies the connection in service selectors and the connection list. | | Host | Settings › AI › Connections | No | Selects the provider address for connections such as LangFuse or a memory service. | | Public Key | Settings › AI › Connections | Yes | Supplies the LangFuse public credential; required for a LangFuse connection. | | Secret Key | Settings › AI › Connections | Yes | Supplies the matching LangFuse secret credential; required for a LangFuse connection. | | Api Key | Settings › AI › Connections | Yes | Supplies the memory provider credential; required for Mem0 and INSIDE Memory Service connections. | | AI Account | Settings › AI › Connections | Yes | Selects the account used by INSIDE Memory Service for profile extraction. | ## Limits and good to know - Connections belong to a website. A connection on one website is not automatically available on another. - **Capabilities** describe supported functions, not current service health. - A passing connection test does not prove that every tool, prompt or model operation will succeed. - Saving a connection does not enable it for all AI features. - Removing a connection can leave existing selections unusable. Arrange replacement selections before removal. - Provider retention and access policies still apply to content sent to an external service. ## Depends on and works with ### This feature uses - **Configured in** [AI feature settings and change history](AI-001.md): Each supported AI feature selects the prepared connections it uses for prompts, tracing or memory. - **Configured in** [AI accounts and shared defaults](AIP-001.md): AI accounts select the shared privacy, guardrail or Salesforce connections used by their features. - **Configured in** [MCP service authentication](AIP-018.md): MCP connection authentication settings determine the credentials sent for discovery and tool calls. - **Works better with** [AI integrations and automation tools](AI-018.md): Activepieces connections expose automation flows for AI functions. - **Works better with** [AI agent service connections](AI-024.md): External agent connections can supply an alternative AI execution service. - **Works better with** [AI Experiments Lab for response testing](AI-032.md): LangFuse connections can provide experiment datasets and results. - **Works better with** [External AI prompts and versions](AIP-014.md): LangFuse connections supply managed prompts. - **Works better with** [AI execution tracing with Langfuse](AIP-015.md): LangFuse connections receive optional AI traces. - **Works better with** [Persistent AI memory connections](AIP-016.md): Mem0 and INSIDE Memory Service connections provide long-term context. - **Works better with** [MCP tool connections for AI](AIP-017.md): MCP Server connections provide remote tools. - **Works better with** [Personal information masking for AI](AIP-021.md): Privacy connections support sensitive-data processing. - **Works better with** [AI guardrail policy connections](AIP-022.md): Amazon Bedrock Guardrails connections provide external guardrail checks. - **Works better with** [Salesforce Models API connection](AIP-027.md): Salesforce AI Connection supplies reusable credentials for Salesforce AI accounts. - **Works better with** [UCP product catalogue connection](COM-005.md): UCP connections discover commerce services and their supported capabilities. ### Used by - **Required by** [External AI prompts and versions](AIP-014.md): External prompts need a configured LangFuse connection. - **Required by** [AI execution tracing with Langfuse](AIP-015.md): External tracing requires a configured LangFuse connection. - **Required by** [Persistent AI memory connections](AIP-016.md): Memory needs a configured Mem0 or INSIDE Memory Service connection. - **Required by** [MCP tool connections for AI](AIP-017.md): Tool discovery and calls need a configured connection. - **Required by** [MCP service authentication](AIP-018.md): Authentication is stored on a website's MCP connection. - **Required by** [Personal information masking for AI](AIP-021.md): Automatic privacy processing needs a selected privacy service and an available analyser. - **Required by** [AI guardrail policy connections](AIP-022.md): External service configuration supplies the policy connection. - **Required by** [UCP product catalogue connection](COM-005.md): An existing registered UCP connection supplies the business capabilities and connection settings. - **Configures** [AI instructions and templates](AI-002.md): Powerfront configures the connections available for external instructions. - **Configures** [AI integrations and automation tools](AI-018.md): MCP and Activepieces functions require their corresponding configured service connection. - **Configures** [Reusable AI skills and instructions](AI-020.md): Powerfront configures the connections available for a skill that uses external instructions. - **Configures** [AI agent service connections](AI-024.md): Powerfront configures the external service connection selected for agent execution. - **Configures** [AI guardrail routing and fallback actions](AI-025.md): Guardrail connection rules take precedence over account rules. - **Configures** [AI Experiments Lab for response testing](AI-032.md): The selected external connection supplies remote datasets and receives experiment results. - **Configures** [AI accounts and shared defaults](AIP-001.md): Account selections refer to configured external service connections. - **Configures** [Salesforce Models API connection](AIP-027.md): A dedicated Salesforce AI connection supplies authentication configuration. ## FAQ and troubleshooting ### How do I add or change a connection? Ask Powerfront to manage **Connections** for your website. Provide the service type, purpose and required connection details. ### Why is my connection absent from a selector? Selectors show connections with the capability needed by that control. Check the website and connection type. Ask Powerfront to confirm the saved connection if it should be available. ### Why does a feature fail even though the connection test passed? The test may cover only basic service access or discovery. Check the selected prompt or tool and the remote permissions required for that operation. Ask Powerfront to check the selected account. ### Can I reuse a connection across features? Yes, within the website, where its type supplies the required capability. Each feature must select it separately.