Last Updated on June 22, 2026 by Holland Rocha
We’ve been heads down building lately, and we have a lot to show for it. Over the past few months, Re:amaze has shipped a new set of AI customer support tools that give you better visibility into how your AI agent works, ways to test it before it talks to a real customer, and a smarter way to train it over time. Here’s a rundown of everything that’s new.
See Which Sources Your AI Agent Is Drawing From
When your AI agent responds to a customer, it’s pulling from the knowledge sources you’ve set up. Now you can see exactly which ones it used for any given response.
A source count appears directly on each AI response in a conversation. Click it and you get the full breakdown, with each source labeled by type. It’s a quick way to spot if the AI leaned on something outdated or pulled from the wrong place.
We also added an Improve Response button to every AI reply. If an answer isn’t quite right, you can submit a correction on the spot, and it gets stored for the AI to reference going forward. All your corrections are easy to review and manage in one place, and you can edit or delete them at any time.
Learn more about AI knowledge sources and corrections
Test Your AI Agent
Once your AI agent is enabled, you can run it through its paces in a dedicated test environment. And honestly, even after you’ve had it running for a while, it’s worth coming back to regularly.
You can send it any question you’d expect a customer to ask, follow up with more messages, and see the full answers it generates in real time. That’s really the point — not just checking whether it responds, but reading the actual responses and deciding if they’re accurate, on-brand, and genuinely helpful. If something looks off, each response shows exactly which knowledge sources the AI pulled from, and you can jump directly into any of them to make updates. Then test again. And again.
It’s a great habit to build any time you add new content to your knowledge base, update a response template, or save a new learned answer. Small changes can have a bigger impact on responses than you’d expect, and the test environment is the best place to find that out.
Learn how to test your AI agent
Train your AI agent from real conversations
When your AI agent can’t answer a customer’s question, it escalates the conversation to a human agent. Once that conversation is marked as Done, Re:amaze automatically generates a suggested answer based on what happened. Those suggestions show up in a review queue for you to approve, edit, or delete. Nothing gets added to your AI agent’s knowledge base until you say so.
Approving a suggestion means your AI agent can reference that answer the next time a similar question comes in. The ones you skip expire after 14 days. You can save up to 50 answers total, so be selective. Prioritize the ones that best reflect your policies, processes, and tone.
Over time it becomes a pretty natural feedback loop — your real support conversations gradually make your AI agent better at handling the next ones.
Learn how to use Learned Answers to train your AI agent
More to Come
These updates are all part of a bigger push to make the AI agent in Re:amaze something you can actually trust and fine-tune over time, not just set up and hope for the best. There’s more in the works, so stay tuned.
Have questions about any of these features? Our Help Center has you covered.