Use Cases

Seven concrete things you can build with botts.ai today, from a customer support widget to a CRM-connected voice agent, a researcher restricted to trusted sources, and a Monday-morning briefing that writes itself, with the exact features to combine.

botts.ai is a general-purpose AI agent platform. The same building blocks, Agents, Knowledge Bases, Deployments, Forms, Integrations, and Scheduled Tasks, combine into very different solutions depending on the channel and configuration you choose.

This page walks through seven concrete setups, each built entirely from features that exist in the product today.

Customer Support Widget

A chat widget on your website that answers product, billing, and how-to questions from your own documentation: 24/7, in your branding.

Features to combine

• Knowledge Base: crawl your FAQ page, help center, and product documentation. Set a crawl schedule (daily, weekly, or monthly) so the agent stays in sync with your docs. See Knowledge Bases.
• Agent with a support-focused System Prompt: define tone, answer length, and escalation rules. See Agents.
• Website deployment: embed the widget with a single script tag; it works on any site. Customize the welcome message, up to 5 suggested messages, the launcher (bubble, bubble with message or card), colors, logo, and the feedback (thumbs) toggle in the widget editor. See Deployments.
• AI voice (optional): in the widget editor's AI voice section, turn on Enable AI voice so visitors can talk to the agent instead of typing, in the same conversation. Pick a Voice, a spoken Welcome message and the Default opening mode. Voice runs on GPT Live 1 in Microsoft Azure, Sweden (processed globally), so it does not stay in Switzerland, and it is billed per minute. See AI voice.
• Backup model: configure a backup model, ideally on a different provider, so the widget keeps answering if the primary model has an outage. See Deployments.
• Support Ticket form (optional): enable the built-in Support Ticket template so the agent can capture an escalation (subject, description, email, priority) when it can't solve the issue. See Lead Capture with Forms below.
• HTTP tools (optional): connect your own API (for example an order-status endpoint) so the agent can answer account-specific questions. Admin or Owner role required to create tools. See Integrations.

How to set it up

1. Open Knowledge in the sidebar, create a Knowledge Base with New Knowledge Base, and add your help-center URL as a website source. Run the first crawl and set a recurring schedule.
2. Create an agent under Agents, write the System Prompt, and link the Knowledge Base.
3. On the agent's Deploy tab, add a Website deployment, pick a model (and a backup model), and customize the widget in the editor with live preview.
4. Copy the embed snippet and paste it into your website.
5. Review real conversations under Conversations and fill Knowledge Base gaps the transcripts reveal. Save the replies that matter as tests with Save as Test on the agent's Chats tab, and press Run All on its Tests page before each change goes live. See Conversations and Testing.

Example System Prompt

You are a support assistant for CloudTech Solutions.
Answer questions about our products, billing, and technical setup
based on the knowledge base.

Rules:
- Be concise. Most answers should be 2-3 sentences.
- If a question requires account-specific information,
  ask the customer to contact support@cloudtech.com.
- If you cannot solve the issue, offer to file a support ticket
  and collect the details with the support_ticket form.
- Never share internal processes or pricing breakdowns.

Internal Knowledge Assistant

A private assistant your team chats with inside the botts.ai dashboard, trained on your employee handbook, IT guides, and policies. Nothing is embedded on a public website.

Features to combine

• Knowledge Base with uploaded documents: upload your internal documents as PDF, DOC, DOCX, TXT, or MD files (up to 100 MB per file). You can also crawl a publicly reachable site such as your documentation portal. See Knowledge Bases.
• Chat (Internal) deployment: deploys the agent to the dashboard's Chat page, where your team members talk to it. One Internal Chat deployment per agent. See Deployments.
• Access control: choose who can use the assistant: Organization (all members), Role-based (minimum role Builder or Admin), or Invite Only (a specific member list). Members with the chat-only Member role see only the agents their access mode permits. See Team & Billing for roles.
• Advanced Knowledge Base Browsing (optional): a toggle under Built-in Tools in the agent editor that lets the agent list Knowledge Base sources and read full documents instead of only searching snippets. Useful when answers require reading a whole policy document.
• Backup model: Internal Chat deployments support a backup model for failover, just like Website deployments.
• File upload (optional): a toggle on the deployment that lets colleagues attach a document, an image or a spreadsheet to the conversation and ask about it, without adding it to the shared Knowledge Base. Right for a one-off contract or report that nobody else needs. See Attaching documents.

How to set it up

1. Create a Knowledge Base and upload your internal documents (drag and drop supports multiple files at once).
2. Create an agent, link the Knowledge Base, and optionally turn on Advanced Knowledge Base Browsing under Built-in Tools.
3. On the Deploy tab, add a Chat (Internal) deployment. Pick the access mode, optionally add a Description (shown below the title when the chat starts), decide whether to allow File upload, and optionally upload a Chat Picture.
4. Invite your team under Organization → Members. Colleagues with the Member role land in Chat as soon as they sign in. Owners, admins and builders open Chat from the sidebar.

Lead Capture with Forms

The agent collects structured data (name, email, booking details, ticket information) in the middle of a natural conversation, validates it, and stores it as a form submission you can review and export.

Forms work on every channel: the website widget, internal chat, and phone calls.

Features to combine

• Form / Lead Capture: a section in the agent editor (Configure tab) with a master toggle. Enable it, then pick built-in templates or define custom forms.
• Email notifications (optional): check Send notification on submission to email each submission to your organization's notification email address (set on the organization settings page).
• Forms tab: every submission appears on the agent's Forms tab, filterable by form name, with CSV export and delete.
• Contact details: name, email and phone from a submission are added to the end user's record (values already stored are kept). The phone number then labels that person's conversations on the agent's Chats tab and can be searched under Conversations. The full submission stays on the Forms tab. See People & Memory.

Built-in form templates

Custom forms

If the templates don't fit, build your own form under Custom Forms: give it a Form Name, then add fields, each with a name, a type (Text, Number or Yes/No) and the Required flag. Names are saved in lowercase with underscores ("Company Name" becomes company_name), and the agent asks for each field by its name, so pick names that say what you need.

How it works in conversation

1. The agent asks for the form's fields conversationally. It doesn't render a literal form; it collects the values in chat (or by voice on a phone call).
2. Submitted data is validated against the form's schema; if something is missing or invalid, the agent asks the user to correct it.
3. Each form can be submitted once per conversation. Duplicate submissions are rejected automatically.
4. The submission is stored and, if enabled, emailed to your notification address.

Phone Agent

A voice agent on a real phone number that answers every call, for example a receptionist that shares opening hours, answers service questions from the Knowledge Base, and takes booking requests with a form.

Features to combine

• Phone deployment: created instantly when you select the Phone channel on the Deploy tab. Phone deployments run on GPT Live 1; you pick one of 8 voices (Gleam, Cedar, Marin, Meridian, Willow, Stone, Vesper, Ripple), with samples on the voices page. See Deployments.
• Phone number: buy a Swiss or US local number into your organization's number pool and assign it to the deployment. Buying and managing numbers requires the Admin or Owner role and counts against your phone-number quota.
• Welcome message: the first thing the agent says when it picks up.
• Knowledge Base: the same Knowledge Base the agent uses in chat also answers questions on calls. See Knowledge Bases.
• Booking Request form (optional): enable the built-in Booking Request template so the agent collects name, email, date, and service by voice. See Lead Capture with Forms.
• Appointment booking (optional): instead of taking a request, the agent books straight into a Cal.com calendar, on the call itself. botts.ai sets the tool up for you for now. See Appointment booking.
• Call settings: the language the call is expected in (Auto, English, Deutsch, Français, Italiano: the greeting and the first turns start in it, a caller who clearly speaks another language is followed), call recording (off by default), and a maximum call duration (dashboard default: 10 minutes).
• Access control (optional): Public is the default for customer-facing lines. For an internal line, choose Organization, Role-based, or Invite Only: the caller's number is matched against the phone numbers in your members' profiles, and unauthorized callers hear a spoken "Access denied" message. See Your account settings for where members add these numbers.

How to set it up

1. On the agent's Deploy tab, select the Phone channel. A phone deployment is created with defaults and opened for editing.
2. Buy or assign a phone number from your organization's pool.
3. Set the welcome message, pick a voice, and adjust call settings.
4. Call the number and test. Transcripts (and recordings, if enabled) appear under Conversations. See Conversations.

Example System Prompt

You are the phone assistant for Dr. Müller's dental practice in Zurich.
Answer calls professionally and helpfully.

Rules:
- Provide information about our services, opening hours, and location.
- For appointment requests, collect the patient's details with the
  booking_request form and confirm we will call them back.
- For emergencies, give the emergency dental number: +41 44 200 33 88.
- Be calm, professional, and reassuring.

CRM-Connected Voice Agent

A phone agent that does more than answer questions: when a customer calls, it asks for their phone or customer number, looks them up in your CRM, greets returning customers by name, and logs the call outcome, or books a follow-up, back in your CRM, all during the conversation. This is the pattern behind "voice agent plus your business systems."

Features to combine

• Phone deployment: a voice agent on a real number. See Deployments.
• A connection to your CRM: the bridge to your business system. If your CRM runs an MCP server, connect it and import the tools you need. If it publishes an OpenAPI spec, point the connection at the spec's URL and pick the endpoints. Otherwise define the endpoints by hand: a lookup_customer (a GET to your CRM's contact endpoint, filtered by phone) and a log_interaction (a POST that adds a note or task). See Integrations.
• Secrets: store your CRM API key once and reference it as $secret:crm_api_key in the tool's authentication, so the key never appears in a prompt or tool definition. See Integrations.
• Knowledge Base (optional): answer product and service questions on the call from your indexed content.
• Booking Request form (optional): capture a callback or appointment by voice when the request can't be resolved live.
• Customer Memory (optional): remember preferences and standing instructions per caller across calls. See People & Memory.

How to set it up

1. In Integrations → Secrets, add your CRM API key as a secret (for example crm_api_key).
2. In Integrations, click Add integration → Build your own and set up an HTTP API connection for your CRM with base URL https://your-crm.example.com, Auth Type Bearer Token and token $secret:crm_api_key, then give it two endpoints: lookup_customer (GET /contacts?phone={phone}, with a phone parameter) and log_interaction (POST /contacts/{id}/notes, with id and note parameters). Test both from the dialog before continuing.
3. Create an agent, attach both tools, and optionally link a Knowledge Base.
4. On the Deploy tab, add a Phone deployment, assign a number, and pick a voice.
5. In the System Prompt, tell the agent when to call each tool: ask for the caller's phone or customer number at the start and look them up, use their name, and log a summary before hanging up.

Example System Prompt

You are the phone assistant for [Your Company].
At the start of every call, ask for the caller's phone number or
customer number and use the lookup_customer tool with it to check
whether they are an existing customer.

Rules:
- If the customer is found, greet them by name and reference their account.
- If not, ask how you can help and offer a callback with the booking_request form.
- Answer product and service questions from the knowledge base.
- Before ending the call, use log_interaction to save a one-line summary.
- Never read account IDs or internal notes aloud.

Research Assistant on Trusted Sources

An internal assistant that researches on the live web, but only within sources you trust: your regulator, a standards body, an industry association, your own site. The point is not that it can search; it is that it cannot wander.

Features to combine

• Web search with "Only these sites": list up to 10 domains the agent may draw on. Everything else is filtered out before the agent sees it. See Web search.
• Knowledge Base (optional): your own material, which the agent prefers because it searches it first.
• Chat (Internal) deployment: the team asks in the dashboard.
• Chat attachments: a colleague can drop a PDF into the conversation and ask how it squares with the guidance. See Attaching documents.

How to set it up

1. Create an agent and write a prompt that says which sources are authoritative and that it must cite them.
2. In Built-in Tools, turn on Web search, choose Only these sites, and add your domains.
3. Deploy it as Chat (Internal) so the team can reach it.

Monday-Morning Briefing

A short summary that arrives before you open the dashboard: what changed in a system you have connected, what the sources you track have published, or a digest built from your own knowledge base. Nobody has to remember to ask.

Features to combine

• A scheduled task: the instruction, the rhythm, and the delivery. See Scheduled Tasks.
• An agent with the knowledge, connections, or web search the summary needs. A task can only report on what its agent can actually fetch, so give it the tool before you give it the schedule. The agent needs a Chat (Internal) deployment you can use; only those agents appear in the wizard.
• Email delivery: the result lands in your inbox. The full version stays in your chat history.

How to set it up

1. Open Scheduled and describe it in one sentence, for example "every Monday at 8, check our suppliers' status pages and summarise anything that changed".
2. Confirm the wizard's reading of the schedule, and under What should run it? pick the agent that has access to the right material (it must be deployed as Chat (Internal)).
3. Choose Email delivery.
4. Use Run it once now before turning it on. You see the actual output and the actual cost.

Scheduled tasks are personal: the one you create is yours, and colleagues who want the same briefing set up their own. Ask for the shortest useful output, because a daily task is thirty runs a month and each one costs credits.

Building Your Own Use Case

These setups share a common pattern:

1. Define the mission: what specific problem should the agent solve?
2. Gather the knowledge: crawl or upload the information it needs into a Knowledge Base.
3. Set the rules: write a System Prompt that says what the agent should do and what it must never do.
4. Choose the channel: the website widget for visitors, Chat (Internal) for your team, Phone for callers. See Deployments.
5. Iterate: review transcripts under Conversations, use Save as Test on the agent's Chats tab for the replies that matter, fill Knowledge Base gaps, refine the System Prompt, and press Run All on the agent's Tests page before each change goes live. See Testing.

Start with one use case, get it working well, and expand from there. Multiple agents can share the same Knowledge Base, so a support widget and an internal assistant can build on the same content.

Last updated on October 28, 2026