Changelog

Phone calls

Experts can now answer and make phone calls.

• Connect your own Twilio account and put its numbers on your experts. See Bring your own Twilio.
• Each expert has a Phone tab and a Call button for outbound calls.
• Calls can be transferred to a person, and outbound calls handle voicemail.
• Hang up a live call from the app.
• Visitors can call an expert’s number from its public page.

See Phone.

Outbound campaigns

Call a list of contacts with an expert. Import contacts, then start, pause, resume or cancel the campaign, and export its results. A workspace do-not-call list keeps numbers out of every campaign. See Campaigns.

Knowledge

• Shared collections: one set of documents that several experts read from.
• FAQs: question-and-answer documents an expert answers from directly.
• Website import: turn a site’s pages into documents, and fetch them again when the site changes.
• Better search: search now matches meaning and exact words together.
• Sources: chat replies say which documents they used.

See Knowledge.

Test suites

Write test conversations for an expert. A simulated caller plays each one, and a judge checks the expert’s answers against your criteria. Run a suite before you publish a change. See Test suites.

Analysis and Analytics

Every finished call, chat and meeting can be analysed: outcome tags, extracted facts, success criteria, a score, sentiment and a summary. Set it up on the expert’s Analysis tab.

The Analytics page shows outcomes, sentiment, response times and cost over time, what is live now, and exports to CSV. See Analytics.

Experts

• Version history: every save is a version. See which settings changed, and restore any version. See Versions.
• Drafts: unsaved edits are recovered if you leave, and you can test them before you save.
• Variables: use {{variables}} in an expert’s instructions and fill them per conversation, from the widget, a public link or the API.
• Call experience: silence check-ins, a maximum call length, end-of-call phrases, how eagerly the expert takes its turn, and background ambience.
• Guardrails: set how strictly an expert stays within its boundaries. See Guardrails.
• Claude Haiku 5.5 is now the default Claude model for voice calls and for chat replies.

Meetings

• An expert now tells people it is an AI when it joins: by voice, in the meeting chat, or both.
• New modes: chat only, or speak only when called by name.
• Experts remember earlier meetings with the same people.
• Transcript lines are named by speaker.

See Meetings.

Conversations

One inbox for meetings, chats, web calls, phone calls and visitors’ conversations, with filters, search across transcripts, multi-select, bulk delete and export. See Conversations.

For developers

• MCP: every expert is now a Model Context Protocol server for Claude, Cursor and other AI tools. See MCP.
• API reference: the full reference is in the app under Developers → API reference, built from an OpenAPI document at GET /v1/openapi.json.
• Provider keys: run your experts on your own OpenAI, Anthropic, Google, Deepgram or Cartesia keys. See Provider keys.
• Workspace export: download your workspace’s data. See Export.
• Alert channels: send alerts to Slack or PagerDuty. See Alerts.
• Ratings: visitors can rate replies and calls on the public page and the widget.