TL;DR: Communicate with Amplitude in pictures and speech, have agent memory persist between conversations and projects, audit agent conversations, and access a bunch of new tools through MCP.

Two new ways to interact with agents, both pretty self-explanatory.

Image upload: Drop a screenshot or chart into the agent chat and ask about it. Want to recreate a chart from a sketch? Or share a competitor’s dashboard and ask how yours compares? Just paste the image instead of trying to describe it.

Voice mode: Talk to the agent instead of typing. This works the way you’d expect: click the voice mode button and start speaking. Useful when you’re on a call or between meetings and want a quick answer without switching context.

Introducing Agent Memory

Organization context and knowledge docs work well for things that are stable and widely shared: your business model, metric definitions, event naming conventions. But some things for your specific workflows don’t fit neatly into a document.

Agent Memory fixes this. When you correct the agent, give it feedback, or reinforce something that worked, it remembers. Those corrections carry forward into future conversations. If newer information conflicts with an older memory, the agent updates on its own.

You don’t have to configure anything. Agent Memory is now active and learning.

Conversation history across projects

Your agent conversations now persist across projects. If you asked a question about retention in one project last week, you can pick that thread back up from a different project today. No re-explaining, no starting from scratch.

Many Amplitude users juggle multiple projects. An insight in one project often leads to a follow-up question in another; now that trail doesn’t break.

Ask about AI Visibility

Global Agent can now understand your brand performance in the context of AI visibility, sentiment, and competitive changes. Answer questions like, "How has our AI visibility trended over the last 90 days compared to trial starts?" and bring together your business performance and AI search performance. This update will be available starting April 29th.

Audit Agent conversations

Many enterprise governance teams need an answer to a basic question: “What are people asking the agent, and what is it saying back?”

The Agent Audit Tab is that answer. Admins can see who’s using agents, what prompts they’re running, and what the agents are returning. It’s the kind of thing that unblocks an org-wide rollout because you can actually show your governance committee what’s happening instead of hand-waving about it.

MCP: New tools, agent skills, and support for mobile

Get more insights in Claude, Cursor, ChatGPT, or any MCP-compatible client with new tools, a skills repository, and MCP support for mobile:

• Our MCP now has tools spanning more of the Amplitude platform. Analyze session replays, debug agent traces, investigate your web vitals, and pull user feedback trends. See what tools are added here
• Turn your AI chief of staff into an analytics expert with our new open-source agent skills marketplace. Find skills for analyzing dashboards, debugging session replays, investigating A/B tests and more.
• Claude’s mobile app can now use MCP to render Amplitude charts directly in your chat, just like on desktop. Check it out here.
  

We'll keep sharing improvements as we continue to make Agents a better way for you to unlock the power of product data with Amplitude. Learn more about Agents here.