AI employees with one clear job each

Keep each team's agents together in a project inside your company workspace.

Start from a template, such as Support desk, Lead qualifier or Saree and clothing store, or a blank agent. Give it a name, a description and instructions in plain language.

On Home, pick what you want help with: get clients and book calls, post on LinkedIn and Instagram, answer every enquiry, or follow up leads. We set up the agent, its tools and its schedule, and a checklist says what's live, what needs you and what isn't available yet.

Choose how smart each agent is: Smart (the low-cost default), Smarter or Smartest. The builder shows about what a reply costs on each. All three run on Anthropic's Claude models.

Choose the tools an agent may use. For each one, decide whether it runs freely, asks first, or asks and then runs by itself after a waiting time.

Cap how many steps an agent may take on one message. At the limit, it stops and hands the task to your team.

Talk to the draft in a test chat where nothing is really sent. Each reply shows what happened, step by step, and what it cost in rupees.

Send an agent up to 3 photos or PDFs with a message, such as a bill or a damaged parcel, in the test chat, Team chat, website chat or the API. Phones can use the camera. Files are kept for 90 days.

Group the messages customers send into test sets with checks, run them in test mode, compare runs, and block publishing until the sets that must pass have passed.

Run a test set on Smart, Smarter and Smartest at once, see each level's score and what a conversation costs, and switch to the cheapest level that passes in one click.

Each morning, a sample of yesterday's real conversations is checked against plain rules, your never-say list and rules of your own. See the passing rate and the rules missed, and get an email when it drops below your bar.

Edit a draft while the published version stays as it is. Every publish is kept as a version, and you can pause and resume a live agent.

If a change made an agent worse, put an earlier version back live in one click. It is saved as a new version, so the history stays complete.

An agent emails a customer who stopped replying, at most twice, checked by you first unless you choose otherwise. A reply of STOP ends them for good.

Agents reply in English, Hindi or Hinglish, whichever the customer writes in, and say they are AI in their first reply.

Leads an agent saves appear in a list for that agent, with leads from test chats marked.

Describe a task in a sentence and get a first draft of the agent to edit, or improve its instructions with AI.