Changelog

What changed in DurbarAI

Changes people can see or use, week by week, newest first. Gaps that remain are on the limits page.

  • Stop button: a halt records why it was pressed, resuming asks for confirmation first, and both the admin and member apps show a halted banner.
  • Approvals: members decide approvals and questions from their own inbox, and an approver can record why they denied an action.
  • Audit: the trace explorer shows each run's identity chain and whether its receipt verified.
  • Admins can export the members directory as CSV; the export is written to the audit log.
  • Knowledge citations now appear in replies on Slack, email and the web widget, scoped to what the reader may open.
  • Chat replies name the answering agent and its version in response headers.
  • OpenTelemetry chat spans record the response model and finish reasons.
  • Admins can revoke a single web-widget visitor.
  • Admins choose which Slack channels an agent may post to, and a person previews a reply before it is sent.
  • The product is now called DurbarAI, and the default agent is Juneli.

  • Organization-wide consent switch for the AI assistant in Studio.
  • Agents get a voice section in Studio, with a reply tier capped by a workspace limit.
  • Each channel can allow a set of languages, detect the one in use, and switch mid-conversation.
  • A plain-language reading mode for people who want simpler replies.

What DurbarAI cannot do yet

Get one email when DurbarAI goes open source

We write once, on the day the repository opens. No newsletter. Ask, and we delete your address.

By submitting you agree to the terms and the privacy notice.