What shipped,and when.

Agent API and webhooks

Put Kuvra's AI agent behind your own interface. The same knowledge base, handoff and reporting as the widget, over HTTP.

• API keys in Settings → Developers, shown once, stored hashed, revocable at any time
• Ask a one-off question, or run full conversations with streamed replies
• Signed webhooks so your server knows when a person takes over, with retries
• An AI mode per conversation: off, until a person joins, or always on

New sign-in and sign-up screens

Sign in, sign up, and password reset were redesigned to match the rest of the product.

Two-factor authentication

Accounts can now require a second step at sign-in.

• Set up an authenticator app from Settings → Access
• A six-digit code step at sign-in once it's on
• Sign out every other session in one click
• Access settings that were saved but never enforced have been removed

Agent tooling in the inbox

A batch of changes aimed at the time between opening a conversation and closing it.

• Contact sidebar with linked company and every past conversation
• Live SLA countdown in the conversation header
• Knowledge base search and insert from the composer
• Macros that carry a reply plus status, priority, assignment, snooze and tags
• Private follow-up reminders with presets, a custom date and notes
• Keyboard shortcuts for close, snooze and saved replies
• Live preview of what the customer is typing

Widget hardening

A review of the public widget surface, and the fixes that came out of it.

• Fixed a stored-XSS vector in message rendering on both the widget and the dashboard
• Domain allowlist enforced when a visitor session is created
• Length caps and empty-message rejection on visitor input
• Retrieved knowledge base content fenced off as untrusted data in the prompt

Retrieval rebuilt

The AI agent's search stopped being a single vector lookup.

• Follow-up questions condensed against the conversation before searching
• Meaning-based and keyword search, merged with reciprocal rank fusion
• A reranking pass over the shortlist before anything reaches the model

Agent identity end to end

Customers can now see who they're talking to, and whether it's a person at all.

• Profile picture upload for agents
• Name and avatar attached to agent replies in the inbox and the widget
• A clear notice in the widget when a person joins the conversation
• Match my website: paste a URL and Kuvra builds the widget palette from it

Search everything with ⌘K

One search box for the whole workspace.

• Conversations, contacts, companies, articles, saved replies and channels
• Matches as you type, and forgives typos
• Jumps straight to any settings page

Triggers and SLA enforcement

Two settings pages that previously stored rules without acting on them now run server-side.

• When → If → Then triggers across five conversation events
• Seven action types including auto-reply and switching the AI off
• Per-priority first-response and resolution targets, checked every minute
• Breach alerts routed to the assignee, respecting notification preferences

Queue management

Tools for teams whose inbox stopped fitting on one screen.

• Conversation tags with a workspace catalogue
• Saved views built from grouped AND/OR conditions
• Server-side search across subjects, messages and contacts
• Collision detection so two agents don't answer the same conversation

Kuvra Copilot, ratings and team threads

An assistant for your team, a way for customers to rate conversations, and somewhere to talk internally.

• Kuvra Copilot: ask about conversations, contacts or reports from anywhere with ⌘J
• Customers rate conversations from one to five stars, with AI and human ratings reported separately
• Internal threads for the team, with replies, reactions and mentions

Knowledge base imports

Bring the documentation you already have instead of retyping it.

• Crawl a documentation site by URL, up to 200 pages
• Change detection so unchanged pages aren't processed again
• Scheduled re-crawls, described in plain English
• Imported pages stay editable as markdown

Composer, notifications and reporting

The rest of the agent-facing surface caught up.

• AI assist: suggest, summarise, rewrite, shorten or expand a draft
• Attachments, emoji, saved replies with / shortcuts, and @mentions on internal notes
• In-app notifications with per-event preferences and quiet hours
• Agent performance reporting and AI usage metering

Theming, integrations and abuse protection

The foundations under the public surface.

• Full widget theming with separate light and dark palettes
• Gmail, Slack, HubSpot, Linear and Shopify connections per workspace
• Rate limiting with progressive IP bans across HTTP and websockets