What Tars 3.0 can do
Tars covers the whole life of a conversation, from the first gambit you draw in the builder to the report you read a week later. The five stages below are the loop your team works in, and each card under them links to the page for that capability.Lay the conversation out as gambits in the builder, with branching, variables, and translated text.
Hand a turn to the model. It follows your instructions, calls tools, and leaves by an exit route.
Turn on Web Search, Code Interpreter, Data Extraction, and File Generator, plus generated interface components.
Import a website, upload documents, and tune retrievers so answers come from your own content.
Connect a toolkit, choose the actions an agent may call, or write a custom action in TypeScript.
Keep an agent on topic, test it before you publish, and control what end users see of its reasoning.
Publish a version, then distribute it as a web widget, an Agent Link, or a screen inside a mobile app.
Connect a WhatsApp number to a deployment and answer end users on the channel they already use.
Human agents take over in the Agent Workspace, with inboxes, assignment, Copilot drafts, and CSAT.
Watch conversations as they happen and read every message, event, tool call, and captured variable.
One persistent profile per end user, with identity verification, segments, consent, import, and export.
Send an outbound WhatsApp campaign to a segment, handle the replies, and pay from the campaign wallet.
Track goals, read the Overview and Team tabs, and open Explore to reach the conversations behind a number.
REST APIs, webhook triggers, the widget JavaScript API, signed identity, and the mobile bridge.
Audit log, SIEM streaming, your own model key, retention windows, legal holds, and erasure requests.