--- title: AI Settings and Controls description: "Manage agent availability, memory, AI context, model providers, connectors, and audit logs from one settings page." product: general token_estimate: 3430 --- # AI Settings and Controls > For AI agents: a documentation index is available at [/docs/llms.txt](/docs/llms.txt). Append `.md` to any page URL for markdown, or send `Accept: text/markdown`. Organization Admins configure Global Agent and Wave availability, AI context, model providers, agent memory, audit logs, and MCP connectors from the _AI Controls_ page, organized across four tabs: **Setup**, **Memory**, **Audit**, and **Connectors**. ## AI Controls settings Access organization-level AI Controls from _Settings > AI Controls_. The _AI Controls_ page includes: - **Agents availability**: Choose an availability state for Global Agent and Wave in your organization. - **AI context**: Provide instructions and upload reference files to help Agents understand your business. - **Shared Cursor integration**: Share a Cursor API key with org members for use in Cursor coding agents. - **Model providers**: Add your own model provider keys for Agent Analytics custom evaluators. - **Memory**: Review and manage organization-level memories Global Agent has learned from conversations. - **Audit**: View Global Agent (Ask AI) and Wave conversation threads across your organization. - **Connectors**: Control which MCP connectors org members can authorize. - **Disable all AI**: Contact Amplitude support to turn off all AI features across your organization. ## Agents availability Control whether Global Agent and Wave are available in your organization. On the **Setup** tab, select one of three states: - **Off**: Turns off Agents for everyone in your organization. - **Admin preview**: Limits Agents to Organization Admins only. Use this state to test Agents before turning them on for your whole organization. - **On for organization**: Turns on Agents for everyone in your organization. Select **View which roles have access** to see which of your organization's roles have access to Agents. If you need to manage access by role instead of by this organization-wide state, select **Looking for role-based access controls?** to go to [Role-based Access Controls (RBAC)](https://amplitude.com/docs/admin/account-management/role-based-access-controls-rbac). > **Note:** > > These settings affect all users in your organization. Individual users can't override the organization-level state. ### Connect Slack and Microsoft Teams The Setup tab also shows whether your organization has connected Global Agent to Slack or Microsoft Teams: - Select **Connect Slack** to let users ask questions by messaging @Amplitude in channels and access Global Agent directly in Slack. - Select **Connect Teams** to let users ask questions by @mentioning Amplitude in chats and channels and access Global Agent directly in Teams. For platform setup details, refer to [Integrate Slack with Amplitude](https://amplitude.com/docs/analytics/integrate-slack) and [Integrate Microsoft Teams with Amplitude](https://amplitude.com/docs/analytics/integrate-microsoft-teams). For how to use Global Agent once it's connected, refer to [Global Agent in Slack, Microsoft Teams, and Discord](https://amplitude.com/docs/amplitude-ai/slack-teams). ## Audit The _Audit_ tab gives organization admins visibility into Global Agent and Wave conversations across the organization. Use this tab to monitor usage and verify that Agents operate as expected. ### View the audit log 1. Navigate to _Settings > AI Controls_. 2. Select the **Audit** tab. 3. Optionally, filter by date range or search for a specific session. 4. Review the table of threads: **Session**, **Agent**, **User ID**, and **Started**. The audit log displays Global Agent (labeled **Ask AI** in the product) and Wave conversation threads from users in your organization. Other agent runs, such as Specialized Agents, Custom Agents, and Agent Analytics sessions, aren't included. Select **Export JSON** to download the current audit log. ### Use cases for audit Use the audit log to: - **Monitor adoption**: Track how many users engage with Global Agent and how often. - **Verify appropriate use**: Ensure users interact with Global Agent according to your organization's guidelines. - **Troubleshoot issues**: Investigate specific conversations when users report problems or unexpected behavior. - **Inform context improvements**: Identify patterns in user questions to improve your AI context configuration. > **Note:** > > Only organization admins can access the Audit tab. ## AI context Provide custom instructions that teach Agents (both Global and Specialized) how your business works. For guidance on writing effective context, refer to [AI context](https://amplitude.com/docs/amplitude-ai/ai-context). ### Two levels of context Agents use two layers of context: Organization and Project. - **Organization context**: Applies to all projects. Use for company-wide standards like business model, KPI definitions, standard terminology, and global filters. Limit: 10,000 characters. - **Project context**: Applies to one project. Use for product-specific events, metrics, segments, and overrides to org defaults. Limit: 10,000 characters. At runtime, Agents receive both context sources. Project context overrides organization context when they conflict. ### Text instructions Enter your context as structured markdown text. Use headings, bullets, and backticks to make it easy for Agents to parse. To add text instructions: 1. Navigate to _Settings > AI Controls_. 2. On the **Setup** tab, find the _Organization Context_ or _Project Context_ card. For Project Context, select the project from the dropdown next to the heading. 3. Select **Edit**. 4. Enter your instructions in the markdown editor, or select **Upload markdown file** to add instructions from a file. 5. Select **Save**. Select **Copy markdown** at any time to copy the current instructions, or **Preview** to see the rendered markdown. ### File uploads Upload reference files for Agents to use alongside your text instructions. Files provide additional context without consuming your 10,000-character instruction limit. Supported file formats: | Type | Formats | Size limit | | --- | --- | --- | | Documents | TXT, MD, HTML, DOC, DOCX, PDF | 25 MB | | Spreadsheets | CSV, XLS, XLSX | 25 MB | The 50-document storage limit applies separately to each scope: you can store up to 50 documents at the organization level and up to 50 documents in each project. Each scope's limit covers both manually uploaded files and any files synced from external sources. To upload a file: 1. Navigate to _Settings > AI Controls_. 2. On the **Setup** tab, find the _Files_ section. 3. Select **Upload file** and choose one or more files to upload. Use file uploads for: - Data dictionaries or event catalogs in spreadsheet format. - Product requirement documents or specifications. - Style guides or documentation standards. Files uploaded to a project also describe that project's tracking plan. Amplitude uses them to write [AI-generated descriptions](https://amplitude.com/docs/data/event-property-descriptions#ai-generated-descriptions) for the project's events and properties, preferring them over its own usage data. Organization-level files don't feed those descriptions. > **Note:** > > File uploads supplement your text instructions; they don't replace them. Use text instructions for rules and definitions, and file uploads for detailed reference material. ## Shared Cursor integration Share a single Cursor API key with everyone in your organization so members can use it with Cursor coding agents. 1. Navigate to _Settings > AI Controls_. 2. On the **Setup** tab, find the _Shared Cursor Integration_ section. 3. Paste your Cursor API key. 4. Select **Connect**. > **Note:** > > A member's personal Cursor API key always takes precedence over the shared organization key. ## Model providers Bring your own model provider key (BYOK) so Agent Analytics can run custom evaluators and calibration runs on your behalf. Amplitude's built-in signals run without any provider key. Add one only if you define your own quality evaluators or run calibration against labeled examples. Amplitude supports the following providers: - OpenAI - Azure OpenAI - Anthropic - Google Gemini - Fireworks AI - Amazon Bedrock - Google Vertex AI To add a provider key: 1. Navigate to _Settings > AI Controls_. 2. On the **Setup** tab, find the _Model Providers_ section. 3. Select **Connect Provider Key** next to the provider you want to add. 4. Enter your key and save. > **Note:** > > Amplitude encrypts keys at rest and doesn't show them again after you save them. Keep your own copy if you need to reference or rotate the key later. For more detail on custom evaluators and calibration runs, refer to [Set up Agent Analytics](https://amplitude.com/docs/amplitude-ai/agent-analytics/setup#running-custom-evaluators). ## Memory The _Memory_ tab shows organization-level memories: facts Global Agent has picked up across conversations, such as shared business context or recurring preferences. Org admins can review and delete these memories. For more detail on how memory works and how to manage it, refer to [Memory in Global Agent](https://amplitude.com/docs/amplitude-ai/global-agent-memory#organization-memory). ## Connectors The _Connectors_ tab controls which MCP connectors org members can authorize for Global Agent, and whether members can add custom MCP servers. - Toggle **Allow custom MCP servers** to let (or block) org members from importing and using custom MCP servers. - For each built-in connector (Atlassian, GitHub, Granola, Linear, Notion, Sentry, and Slack), use the **Enabled** toggle to control whether org members can connect it, or select **Edit** or **Delete** to manage it. - Select **Add custom MCP server** to add a connector beyond the built-in list. For full detail on supported connectors and how to manage them, refer to [MCP Connectors for Global Agent](https://amplitude.com/docs/amplitude-ai/global-agent-connectors#manage-available-connectors). ## Disable all AI To disable Global Agent for your organization, select **Off** in the [Agents availability](#agents-availability) section. You can also toggle other individual AI features on or off from _Settings > AI Controls_. To disable all Amplitude AI features across your entire organization, contact [Amplitude Support](https://amplitude.com/contact-us). Amplitude processes opt-out requests and disables all AI functionality for your organization. > **Note:** > > Disabling individual features from the settings page doesn't turn off all AI. For a complete AI opt-out, contact Amplitude Support. ## Frequently asked questions #### Who can access AI Controls? Only organization administrators can access and modify AI Controls. This includes Agents availability, AI context, model provider settings, agent memory, audit logs, connectors, and the option to disable all AI. Individual users can't change these settings. #### What does Admin preview do? Admin preview limits Global Agent and Wave to Organization Admins only. Other users in your organization can't see or use them. Use this state to test Agents before selecting **On for organization** to roll them out more broadly. #### What can I see in the Audit tab? The Audit tab displays a table of Global Agent (labeled Ask AI in the product) and Wave conversation threads across your organization, showing the session, agent, user, and start time for each. Other agent runs, such as Specialized Agents and Custom Agents, aren't included. You can filter by date range, search for a session, or select Export JSON to download the log. Use this information to monitor adoption, verify appropriate use, and identify opportunities to improve your AI context configuration. #### What's the character limit for context? 10,000 characters each for org context and project context (20,000 total). This is roughly 1,500 to 2,000 words, plenty for most use cases. Amplitude's editor recommends 1,000 to 3,000 characters per scope as a practical range. If you're hitting the limit, you're probably including too much detail. Focus on definitions and rules that apply to the majority of questions, not edge cases. #### What file types can I upload as context? You can upload documents (TXT, MD, HTML, DOC, DOCX, PDF up to 25 MB) and spreadsheets (CSV, XLS, XLSX up to 25 MB). Use file uploads for detailed reference material like data dictionaries or product specs that would exceed the 10,000-character instruction limit. #### How many files can my organization upload? The 50-document limit applies separately to each scope. You can store up to 50 documents at the organization level and up to 50 documents in each project. Each scope's limit covers both manually uploaded files and any files synced from external sources. If you reach the limit, remove unused files before uploading new ones. #### What are Model Providers used for? Model Providers let you bring your own model provider key (BYOK) so Agent Analytics can run custom evaluators and calibration runs on your behalf. Amplitude's built-in signals run without any provider key. Amplitude encrypts keys at rest and doesn't show them again after you save them. For more detail, refer to [Set up Agent Analytics](https://amplitude.com/docs/amplitude-ai/agent-analytics/setup#running-custom-evaluators). #### How do I re-enable AI after disabling it? An organization administrator can re-enable Agents by navigating to _Settings > AI Controls_, opening the **Setup** tab, and selecting **On for organization** in Agents availability. If you requested a full AI opt-out through Amplitude Support, contact support again to re-enable AI features for your organization.