This new Ari release allows you to test Ari before it talks to your customers, see the proof behind every answer, shape how it talks and when it escalates, and trust it to know when to leave a thread to a human.

Try it before your customers do.

Ari now has a Playground: a private chat where you can ask questions and see the exact reasoning behind every answer, no real customers involved. When you're ready, run Shadow Mode on your live threads. Ari drafts replies to real customers, but nothing sends, only your team sees it. Once you trust Ari, you can enable it to resolve real customer threads.

See where every answer came from. 

Every reply records the knowledge sources behind it, in the thread, through the API, and via the Plain MCP. You can audit any answer long after it is sent, and Ari links customers to the full help center article inline.

Shape Ari around your team. 

Custom instructions is a free-text box in Ari preferences: write your introductions, escalation rules, product vocabulary, and sign-offs without engineering work. Thread labels and the thread fields you pick now feed Ari too. Together, Ari sounds like your team, escalates on your rules, and already knows which product and plan a request is about.

Ari can respond to more of your customers, and uses judgement to know when not to.

Ari’s search weighs exact wording alongside meaning, and it reads the files your customers attach. Error codes and API names retrieve correctly, and a screenshot or log file is enough to answer from. The tickets that arrive as an image get answered on first touch. Ari also checks the thread first: if a teammate's already replying, it hands off instead of jumping in, and it won't offer to do something it can't actually do.

Learn more →