FAQ

Frequently asked questions about botts.ai. Get answers about pricing and credits, data residency in Switzerland, roles, model failover, channels, languages, and memory.

Getting Started

How long does it take to set up an AI agent?

Most users have their first agent live in under 15 minutes. The process involves three steps: creating a Knowledge Base (adding your website URL or uploading a document), configuring an Agent (writing a System Prompt), and deploying it (embedding a chat widget, assigning a phone number, or chatting internally). See the Quick Start Guide for a step-by-step walkthrough.

Do I need any technical skills to use botts.ai?

No. botts.ai is designed for business users, not developers. Everything, from building your Knowledge Base to configuring your agent and deploying it, is done through a visual dashboard. No coding is required.

To embed the chat widget on your website, you paste a single script tag into your site. If you use a website builder like WordPress, Shopify, or Squarespace, this usually takes less than 2 minutes.

How do I sign in?

With an email address and password, or with Continue with Google on both the sign-up and login pages. If you already have a password account, sign in with your password: an existing email address is not silently taken over by a Google login. After you add a passkey under Settings → Passkeys, you can also sign in with Passkey (Face ID, Touch ID or a hardware key). For extra protection, turn on Two-Factor Authentication in Settings and confirm each sign-in with an authenticator app.

Is there a desktop app?

Yes, of a kind. The dashboard installs as a desktop app from Chrome or Edge (Install app in the sidebar or the header menu): its own window, a dock or taskbar icon, and updates that arrive on their own. It is the same dashboard, so nothing is stored differently and there is nothing extra to manage. The installed app opens on Chat.

Can I try botts.ai before committing to a paid plan?

Yes. The Personal plan starts with a 1-month free trial and the Business plan with a 14-day free trial, with no credit card required and one free trial per person. Trials include a starter credit allowance (150 credits on Personal, 1,500 on Business), enough to build a Knowledge Base, configure an agent, and talk to it. On Business, the first 150 credits are usable right away and the rest unlock as soon as you add a payment method. If you haven't added a payment method by the end of the trial, your workspace is locked until you choose a plan. Nothing is deleted. If you would rather start right away, choose Pay now instead, without the trial in the plan picker; the trial stays unused.

You can also try an agent before creating any account: type your website address into the chat on the botts.ai homepage and a demo agent built from your site is ready in a minute or two. Signing up afterwards keeps that demo as your first agent. Because a website agent needs the Business plan, this path offers Business, with its 14-day free trial, or Business Pro. See the Quick Start.

Knowledge Base

What types of content can I add to a Knowledge Base?

You can add two types of sources:

• Websites: Provide a URL, and botts.ai crawls and indexes the pages automatically. Crawls can run manually or on a daily, weekly, or monthly schedule.
• Files: Upload PDF, DOC, DOCX, TXT, or Markdown files (up to 100 MB per file).

How many pages can my Knowledge Bases hold?

A page is 2,000 characters of extracted text. The cap applies organization-wide, across all Knowledge Bases, and counts crawled and uploaded content combined:

Usage banners on the Knowledge pages show how much of your page and upload-storage allowance you have used.

How often should I update my Knowledge Base?

Whenever your content changes. Each website source has its own crawl schedule (manual, daily, weekly, or monthly), so a frequently changing site can re-index itself automatically. For uploaded files, upload the new version and delete the old one.

Does the agent respond only from the Knowledge Base?

That's up to you. When an agent has linked Knowledge Bases, a knowledge search tool is enabled automatically and the agent uses it to ground its answers. Two other sources can come into play:

• The underlying AI model's own general knowledge, which is always there.
• Web search, if you switch it on for that agent. It's off by default.

If you want the agent to answer only from your content, leave web search off and say so explicitly in the System Prompt (for example: "Only answer using information from the knowledge base. If you cannot find the answer, say so.").

Agents

How does the System Prompt work?

The System Prompt is a set of instructions written in plain language that tells your agent how to behave. It defines the agent's personality, tone, rules, and boundaries. The AI model reads these instructions at the beginning of every conversation and follows them throughout.

Can I have multiple agents?

On Business and higher plans, yes. The number of active agents is limited by your plan: Personal includes 1 agent, Business 2, Business Pro 5, and Enterprise is unlimited. Creating an agent beyond your limit is blocked until you upgrade. Within your limit, each agent has its own System Prompt, Knowledge Bases, tools, and deployments. This is useful when you need different agents for different purposes, for example, one for customer support and another for sales. The number of active deployments is also limited by your plan.

Can one agent use multiple Knowledge Bases?

Yes. An agent can be linked to multiple Knowledge Bases, and one Knowledge Base can be shared by multiple agents. When answering a question, the agent searches all linked Knowledge Bases and uses the best results across them.

How does memory work?

Customer Memory is an opt-in, per-agent feature (off by default). When enabled in the agent's Customer Memory section:

• After a conversation ends, an extraction step identifies persistent facts about the person (name, preferences, standing instructions) and updates a short rolling summary of past conversations.
• Memory is stored per person, per agent, and is loaded into the agent's context at the start of that person's next conversation, so the agent can pick up where it left off.
• A retention period controls how long facts are kept: Keep indefinitely (the default), or 30, 60, 90, 180 days, or 1 year. Expired memories are pruned automatically.
• You can review, edit, or delete individual facts, or wipe a person's memory entirely, from a person's profile under the agent's Chats tab, Grouped view. See People & Memory.

Chat memory is a separate memory for the dashboard Chat. It is on by default: your regular chats build up private facts about you (your role, your projects, how you like answers) plus a short summary of recent work, and every model you pick uses it. Only you can see it. Under Settings → Chat memory you can switch off Remember things about me, remove single facts with Forget this, or clear everything with Forget everything.

Memory extraction runs on Swiss-hosted infrastructure, like the rest of the platform.

Can my agent search the web?

Yes, if you turn on Web search for that agent. It's off by default. The agent searches the live web, opens the pages it needs to read in full, and cites its sources. You can restrict it to a list of up to 10 domains, or exclude specific domains.

Web search sends search queries to a French provider (Staan.ai). A few other features also send data outside Switzerland: phone calls go through a US telephony provider, AI voice (phone deployments and widgets with AI voice) runs on Microsoft Azure in Sweden with global processing, and appointment booking sends bookings to Cal.com in the US. The deployment's Data Assessment panel lists each one that applies. See Web search.

Can an agent run on a schedule, without anyone asking?

Yes. A scheduled task runs an agent on a timer, daily, weekly, or monthly, or once on a date you choose, and delivers the result by email or into your chat. Tasks are personal to whoever created them. See Scheduled Tasks.

Can I connect an existing tool without configuring endpoints by hand?

Usually. If the vendor runs an MCP server, connect it and botts.ai discovers the available tools for you; a catalogue of about two dozen vendors pre-fills the details. If the API publishes an OpenAPI spec, point botts.ai at the spec's URL and pick the endpoints you want. Hand-written endpoints remain available for everything else. See Integrations.

How do I know whether a change made my agent better or worse?

Save the conversations that matter as tests, each with a short description of what a good reply must do, and click Run All on the agent's Tests page (Testing group) before and after the change. Every reply is graded Passed, Failed or Couldn't assess against your description, and History keeps each run, so you can compare a prompt, model or tool change on the same scenarios instead of re-reading chats by hand. Predictions proposes likely customer questions from recent conversations to seed the list. See Testing.

Deployments & Channels

Which channels are supported?

Three channels are available today:

On the Personal plan only Chat (Internal) is available; Website and Phone need Business or higher.

Telegram, WhatsApp, and Email appear in the channel selector but are marked "Coming Soon" and cannot be configured yet. All deployments of an agent share the same configuration and Knowledge Bases; only the channel and model settings differ. See Deployments.

Can I customize the chat widget's appearance?

Yes. The widget editor lets you customize colors and fonts for the header, chat area, message bubbles, and footer, plus the welcome message, suggested messages, logo, bot name, widget position, and size. You also pick a launcher style (Bubble, Bubble with message, or Card with buttons), the widget's language, and optionally turn on AI voice. A live preview shows your changes in real time; click Save Changes to publish them.

What is the backup model, and when does failover happen?

Website and Chat (Internal) deployments can set an optional backup model in addition to the primary AI model. If the primary model's provider has an outage (server errors), rate-limits the request, times out, sends no first word within 30 seconds, or reports the model as unavailable, the request automatically retries on the backup model, so your agent keeps answering during provider incidents. An empty reply is retried once on the same model, then goes to the backup. A model that fails several times in a row is skipped for a while: requests go straight to the backup, or, without one, stop quickly with an error instead of hanging (in the dashboard Chat: "The selected model isn't responding right now").

Failover never happens for ordinary request errors or after a tool with side effects (such as a form submission or HTTP call) has run. In a live chat it also stops once the agent has started streaming a reply. Scheduled task runs are the exception: nobody is reading along, so they can switch to the backup in the middle of a reply. For best reliability, choose a backup model hosted by a different provider than the primary: the form warns you if both are on the same provider. Phone deployments use voice models and have no backup model.

Which countries are supported for phone numbers?

In the dashboard you can buy local numbers in Switzerland and the United States, and pick an area code to get a number for a particular region. Numbers belong to your organization's pool and can be reassigned between phone deployments. Each organization has a phone-number quota: 1 number on Business, 2 on Business Pro, custom on Enterprise. See Phone numbers.

Which languages do agents speak?

The AI models are multilingual: agents respond in the language the user writes or speaks, and you can pin a language in the System Prompt (for example, "Always answer in German").

For phone deployments, the Language setting (Auto, English, Deutsch, Français, Italiano) is the language the call is expected in: the greeting and the first turns start in it, and the agent still follows a caller who clearly speaks another language. The dashboard itself is available in English, German, and Italian. The website widget's own buttons and labels are available in English, German, French, and Italian: its Language setting follows the page the widget is on, then the visitor's browser, or can be fixed to one language. Your welcome message and suggested messages are shown as you wrote them.

Team & Roles

What roles exist?

botts.ai uses four roles per organization:

What can chat-only Members access?

Members only see agents that have an active Chat (Internal) deployment they are allowed to use. Each internal chat deployment has an access mode: open to the whole organization, restricted by minimum role, or invite-only with an explicit member list. Phone deployments can use the same access modes, enforced by caller ID against the Work Phone and Personal Phone members enter under Settings → Profile (see Your account settings).

Data & Privacy

Where is my data stored?

In Switzerland. The platform runs on Swiss infrastructure (Infomaniak), and this applies to everyone regardless of plan or settings:

• Knowledge bases: uploaded files and crawled documents are stored in Swiss object storage; the search index lives in a Swiss database. Embeddings and reranking also run on Swiss infrastructure.
• Conversations and memory: chat storage is in Switzerland.
• Secrets: API keys you store for integrations are encrypted at rest and never leave Swiss infrastructure.

All connections use TLS encryption in transit.

Where do AI models process my conversations?

That depends on the models you choose. Each organization has a Server Location setting (Organization → General) that controls which models appear in the pickers, from Global (default, all models) to stricter tiers such as Swiss-hosted models only or Swiss-hosted models operated by Swiss companies. Every deployment shows a Data Assessment badge listing each country where data is processed: primary model, backup model, and platform infrastructure (always Switzerland). Depending on the setup it also lists the phone provider (USA), the voice model (Sweden, Microsoft Azure, processed globally) for phone deployments and widgets with AI voice, web search (France), and appointment booking (USA, Cal.com). See the data residency blog post for details.

Does botts.ai train AI models on my data?

No. botts.ai does not train models on customer data.

Is botts.ai GDPR friendly?

botts.ai is built with European data protection in mind: Swiss hosting, encryption in transit and at rest, no training on your data, and controls that help you comply:

• Memory retention: set a retention period so customer facts are deleted automatically.
• Deletion requests: delete individual remembered facts or a person's entire memory from the agent's Chats tab (Grouped view); deleting knowledge sources permanently removes their indexed content.
• Model residency: restrict AI processing to Switzerland or Europe with the Server Location setting.

Who can see my conversations?

Only members of your own organization, according to their role. Conversations are never shared between organizations. botts.ai staff do not access your data except for troubleshooting with your permission.

Can I disable conversation logging?

Conversations are stored so you can review them in the dashboard. For phone deployments, Record Calls is a separate, per-deployment toggle that is off by default. Recordings are kept by the US telephony provider and played back through the dashboard.

Can I get my conversations out of botts.ai?

Yes. Export on the Conversations hub, or on a single agent's Chats tab, downloads the full history as JSON, CSV, or Excel, either one row per message or one row per conversation. Internal chats you are not permitted to read export as metadata only. See Exporting conversations.

Can I upload a document straight into a chat?

Yes, in the dashboard Chat. Attach a PDF, Word file, text, Markdown, HTML, or JSON file (up to 25 MB) and ask about it. You can also attach images (PNG, JPEG, WebP or GIF, up to 10 MB; models with the Images badge can see them) and CSV or Excel files, which the agent analyses with code and answers with charts and downloadable files. For agent chats the deployment needs File upload switched on. How long attachments are kept is an organization-wide setting under Organization → General → Document Retention. See Attaching documents.

Technical

Which browsers does the chat widget support?

The chat widget works on all modern browsers: Chrome, Firefox, Safari, and Edge. It is fully responsive and works on both desktop and mobile devices.

Does the chat widget slow down my website?

No. The widget script loads asynchronously, so it doesn't block your page from rendering, and it can be embedded on any domain.

Which AI models does botts.ai use?

botts.ai offers models from multiple providers, including models hosted in Switzerland. The available list evolves over time; the model picker in each deployment shows every model's provider, hosting country, speed and intelligence ratings, and credit rates, filtered by your organization's Server Location setting. You choose a model per deployment and can change it at any time with no downtime; in the dashboard Chat you can also switch models in the middle of a conversation. Voice (phone deployments and the widget's AI voice) runs on GPT Live 1; text channels use text models.

Billing & Credits

How does pricing work?

botts.ai uses subscription plans plus credits. Each plan includes a monthly credit allowance and sets resource limits (knowledge-base pages, upload storage, active agents, active deployments, members):

Personal (19 CHF/month) starts with a 1-month free trial and Business (169 CHF/month) with a 14-day free trial. The legacy Free and Lite plans are no longer offered to new customers; existing Lite organizations keep their plan.

All usage is billed in credits: agent responses (based on the model and the amount of text; the model picker shows each model's rates per 1,000 words, and voice per minute), phone call minutes, website crawling (1 credit per page), document indexing, web search (0.10 credits per search, plus 1 credit per page the agent opens in full), and code runs for spreadsheet analysis (0.5 credits each). Visit the Pricing page for plan prices.

For a rough monthly estimate before you build, the Cost Calculator turns a model, your setup and your expected conversations or calls per month into credits and suggests a plan. The Available AI Models page lists every model's credit rates. Where a model has a separate Cached rate, input the provider serves from its cache is billed at that rate and shows as AI Cached on the Usage page.

Can I buy extra credits?

Yes. On paid plans you can buy credit packages of 1,000 up to 100,000 credits, or enable auto-recharge so credits top up automatically. Plan credits expire at the end of each billing cycle; purchased credits stay valid for 12 months. Plan credits are spent first; after that, purchased credits are spent from the earliest-expiring block.

Which payment methods can I use?

Credit and debit cards, and Link (Stripe's saved-payment service). Subscriptions renew automatically and auto-recharge charges your saved method on its own, so checkout only offers methods that can be charged without you present. TWINT is therefore not available. Prices exclude VAT: checkout adds VAT based on your billing address where it applies, and a business can enter its VAT number.

What happens when my credits reach zero?

Your agents pause until credits are available again:

• The website widget and internal chat stop answering with an "insufficient credits" error, and the widget's AI voice does not start.
• Phone calls are answered with a short spoken message and then end.
• New crawls and file uploads are blocked, and scheduled crawls and scheduled task runs are skipped.

Nothing is deleted: your agents, Knowledge Bases, and conversations remain intact. Service resumes as soon as you buy credits or your monthly allowance renews.

Can I change my plan at any time?

Yes. You can upgrade at any time from the Billing section of your organization settings, and upgrades apply instantly. Downgrading Business Pro to Business takes effect at the end of the billing period. Cancelling also takes effect at period end; the workspace is then locked until you choose a plan again. Nothing is deleted.

What happens if I exceed my plan limits?

Limits block new additions, not existing usage:

• Pages or upload storage: new crawls and uploads are blocked until you delete content or upgrade; existing knowledge keeps working.
• Active deployments: creating another active deployment is blocked; deactivating one frees up the slot.
• Phone numbers: buying a number beyond your quota is blocked; contact support to increase the quota.

Troubleshooting

Why won't my PDF upload, or why is it empty after uploading?

A few common causes:

• Scanned or image-only PDF. botts.ai indexes text. A PDF that is just scanned images has no extractable text, so it uploads but shows as Skipped and adds nothing searchable. Use a text-based PDF, or copy the text into a Markdown file.
• File too large. The limit is 100 MB per file; larger files are rejected before the upload starts.
• Out of storage or credits. Uploads are blocked when you've hit your plan's upload-storage limit or your credit balance is zero. Check the usage banners at the top of the Knowledge Bases page.

Supported types are PDF, DOC, DOCX, TXT, and Markdown. See Knowledge Base.

Why did my agent stop responding to customers?

The most common cause is running out of credits: the website widget and internal chat return an "insufficient credits" message and phone calls end with a short spoken notice. Buy a credit package or wait for your monthly allowance to renew (enabling auto-recharge avoids this). Also check that the deployment is still active: downgrading your plan can deactivate deployments that exceed the new plan's limit. See Team & Billing.

Why are my agent's answers outdated or missing recent content?

An agent only knows what's indexed in its linked Knowledge Bases. After your website or documents change, re-index: click the play button to re-crawl a website source (or set a daily/weekly/monthly schedule), and re-upload changed files. Confirm each source shows indexed, not crawling, failed, or skipped. See Knowledge Base.

My agent gave a wrong answer: how do I improve it?

Three levers, in order: tighten the System Prompt (add explicit rules, and an instruction to answer only from the knowledge base if you want that); make sure the Knowledge Base actually contains the answer (test it with the built-in AI Search); then re-test in the Test Agent panel before redeploying. See Agents.

Why didn't I get my verification code?

Signing up with an email address and password sends a 6-digit code to that address; enter it on the Verify your email screen. The code is valid for 15 minutes. Under Didn't receive the code?, Resend code becomes available after 60 seconds and sends a new code that replaces the earlier one, so use the newest email. Check your spam folder and the spelling of the address. Disposable inbox addresses are not accepted at signup.

Last updated on October 7, 2026