Beam
Open Beam
Guides

The Assistant

Current documentation

A practical, safe guide for owners and teams to train one Beam assistant for text and email before it speaks to a real lead.

The Assistant tab where a workspace owner connects an AI key, builds business context, and safely tests responses
Start hereOne owner sets up and approves the assistant. Teammates use the inbox, take over when needed, and coach it from a real conversation. The assistant starts off, and nothing in the Playground is ever sent.

Who does what

PersonWhat they do
Workspace ownerConnects the AI account, adds business context, chooses the calendar and safety policy, tests it, then approves live behavior.
Team memberWorks conversations in Inbox, takes over when context or judgement is needed, and adds coaching feedback when a reply needs improvement.
ClientReviews the assistant's tone and business facts in their own workspace. They never need access to another client's data or your AI account.

Setup, in the order that works

1. Connect your OpenRouter account

In Assistant, paste an OpenRouter API key and select a model. The key is used server-side for your workspace and is never shown back to your team. OpenRouter usage remains on your own account.

2. Explain the business like you would to a new hire

Describe what you sell, who buys it, typical price range, what makes a lead a fit, what happens next, and what must never be promised. Beam creates a starting playbook: a persona, qualification questions, common objections, and a booking approach. Review it before you go live.

3. Add owner rules in plain language

Use House Rules for business-specific corrections such as “Do not quote pricing until you know what service they need” or “Hand off all cancellation requests.” Rules apply to both text and email. Test every important rule in the Playground.

4. Pick the calendar and handoff conditions

Choose the real calendar the assistant may use. Add the situations where it must stop and hand a conversation to a person, such as legal questions, refunds, unusual pricing, or an angry customer.

5. Test before enabling

Use the Playground's Text and Email tabs. Try a price question, a request for a human, a booking request, and an objection. It is a rehearsal only: it cannot send, queue, or create an appointment.

6. Start with drafts

For email, choose Draft only first. Review a few real drafts in Inbox. Enable live assistant behavior only after the owner is satisfied with the tests and the team knows when to take over.

How the Playground works

The Playground uses the workspace's actual playbook and active House Rules. Choose Text to see message bubbles or Email to see a threaded email draft. Type as a customer or start from a common scenario. Each simulated inbound can produce at most one assistant reply, and the result is never delivered to a contact.

A simple approval testBefore enabling live behavior, run these four prompts in both channels: “What does this cost?”, “Can you call me tomorrow?”, “I need a human”, and “Please stop.” The assistant should answer facts it knows, avoid invented commitments, hand off the human request, and respect the stop request.

How it sells

Real calendar booking

The Assistant tab: AI connected, and the Booking calendar card set to a real CRM calendar with Books real appointments on

Pick a booking calendar in the Assistant tab (any calendar in your connected CRM: round-robin, team, or personal) and the assistant becomes a true setter:

Without a calendar selected, the assistant works in tag-only mode: it confirms a window in words and tags bot-booked-call so your own workflow does the booking.

Follow-ups that do not annoy people

Automated follow-ups are off by default. An owner must deliberately enable them and choose up to three delays in hours. A reply, opt-out, handoff, or human takeover stops the sequence. Beam never treats an uncertain delivery state as permission to send the same message again.

Daily team workflow

  1. Work new conversations from Inbox.
  2. If a reply needs judgement, click takeover and reply as a human. The assistant stops on that thread.
  3. If the assistant's behavior needs changing, use Coach your assistant in the conversation or add a House Rule in Assistant.
  4. Review assistant drafts before sending until the workspace owner is comfortable with the results.
  5. Use the assistant only for contacts with appropriate consent and always honor opt-outs.
Do not turn on live replies to “see what happens”Use the Playground and draft mode first. If you see a duplicate, a missing message, an unclear delivery result, or a reply that needs a human, pause the assistant and resolve the issue before continuing.

When it books

The appointment is on your calendar, the contact gets tagged bot-booked-call in your CRM with the agreed window in the notes, your team channel gets a notification if alerts are on, and an assistant.booked event fires to your webhooks. The loop from first text to booked call runs without a human touching it, and every message is in the inbox if you want to watch it work.