Connect context to Wave
The quality of the opportunities Amplitude discovers depends on the context connected to each Product Area. Discovery agents analyze your connected insight sources, corroborate findings across them, and ground every opportunity in specific evidence. More relevant signal produces sharper, better-evidenced opportunities.
You don't need every source on day one. Start with what you have, and add sources as activity grows. More relevant sources give agents more signal to analyze.
Insight sources
Open a Product Area and go to Configure > Sources. Enable the sources you want Wave to use, then configure their scope and source-specific instructions.
| Source | What it contributes |
|---|---|
| Amplitude Agents | Findings from Specialized Agents, such as the Dashboard, Session Replay, and Customer Feedback agents, and from your custom agents. |
| Analytics | Trends and metric deviations from charts, dashboards, and Autocapture, including conversion rates, volume trends, and segment behavior that form the quantitative baseline. |
| Session Replays | Visual evidence of how users behave: friction signals like rage clicks and dead clicks, and where users hesitate or abandon. |
| AI Feedback | Themes from qualitative feedback: complaints, feature requests, and praise drawn from surveys, support tickets, and other channels. |
| Agent Analytics | Trace clusters, failure patterns, and quality trends from AI agents and AI-powered product experiences. |
| Competitive Analysis | Benchmarks and competitor signals, used for inspiration and market-aware recommendations. |
| PR Reviews | Patterns and recurring feedback from pull request reviews in your connected repositories. |
| Experiments | Results from past experiments, turned into rollout or product improvement ideas. |
| Web Vitals | Performance signals collected through Autocapture, such as latency and page experience, that help identify technical friction. |
| External sources | MCP connectors that let discovery analyze proprietary data sources and internal tools. Only the Product Area owner can add them. |
Wave can also take signal from Slack conversations. To set that up, go to Connect Slack.
Sources adapt to your setup
Discovery uses the sources available in your project and Product Area. For example, Session Replay contributes signal when replay is active, and experiment results contribute signal when you run experiments. You can edit Product Area settings to refine product understanding, investigation guidance, plan generation, and coding-agent handoff.
Scope source inputs
Select Configure beside the source you want to configure:
- Analytics: Select dashboards and charts, and add analytics data filters.
- Session Replays: Set the relevant events or funnel context for replay investigations.
- AI Feedback: Set keywords and the lookback window for relevant feedback.
- Competitive Analysis: Add competitors or comparable products and investigation guidance.
- Web Vitals: Limit performance signals to the relevant product context.
Save the source configuration. Keep ongoing source instructions focused on the Product Area. For a temporary request, use a question for a single investigation run.
Connect external MCP sources
External MCP sources let Wave investigate connected systems without attaching a custom agent to the Product Area.
- Open Manage MCP Connectors from Sources to configure connections in Agent Connectors.
- Connect the external tool using your organization's approved access.
- Return to the Product Area's Sources section. The Product Area owner can enable the connector and add instructions for how Wave should use it.
- If a tool shows Needs Reconnect, reconnect the tool before using its evidence.
A connector supplies additional context; it doesn't replace the Product Area's boundaries or your permissions in the external system.
Custom-agent attachments
Attaching a custom agent directly to a Product Area has limited availability. Custom-agent attachments are separate from external MCP sources. Contact your Amplitude representative if you need to attach a custom agent. Don't assume a published custom agent automatically participates in Wave investigations.
Connect your codebase
The highest-leverage context source is your codebase. When you connect GitHub repositories to a Product Area, discovery and execution improve in three ways:
- Product understanding: Amplitude indexes how the Product Area works, what the measurement plan looks like, and even who built different features, which improves discovery quality.
- Grounded plans: Recommendations reflect the current state of your code, so an opportunity can point to where a change should land instead of describing it abstractly.
- Agent-ready execution: A connected repository lets coding agents draft pull requests directly from an opportunity's execution plan.
To open Wave settings, click Settings in the Wave navigation. Then:
- In Wave settings > Codebase, connect the GitHub integration. GitLab and a GitHub access token are also available.
- In Wave settings > Coding Agents, add a key for Cursor, OpenAI, or Anthropic.
- In Wave settings > Repository Index, choose the repositories Wave indexes.
Start with a small set of repositories so recommendations reflect the current state of the product and where a change should happen.
Access and privacy
Opportunities operates within your existing Amplitude environment and honors your organization's data access controls and permissions. Amplitude uses connected code to ground recommendations for your organization, not to train shared models. For how Amplitude secures code indexing and coding-agent sandboxes, go to Amplitude coding agent.
How context becomes evidence
Discovery doesn't just read these sources: it cites them. When an opportunity makes a claim, that claim links back to the specific chart, dashboard, session replay, or feedback theme it came from. Evidence strength varies: exploratory opportunities state a hypothesis and validation plan, while stronger claims need stronger evidence.
To learn how evidence and citations appear on an opportunity, go to Opportunities.
Was this helpful?
For AI agents: a documentation index is available at /docs/llms.txt. Append .md to any page URL for markdown, or send Accept: text/markdown.