Reference

A single-page lookup for botts.ai — plan limits, what costs credits, supported file types, channels, data-residency tiers, the widget embed, HTTP tools and secrets, form fields, and an explicit list of what the platform does not do (yet).

Quick-reference tables that pull the hard numbers and rules together in one place. For the "why" and "how," follow the links into the full guides.

Plans and limits

New workspaces choose a plan right after signup. Personal starts with a 1-month free trial and Business with a 14-day free trial — no credit card required, one free trial per person. Prices are the same number in CHF, EUR, and USD; paying yearly is about 17% cheaper. Enterprise is custom and arranged with sales. The legacy Free and Lite plans are no longer offered to new customers; existing Lite organizations keep their plan.

PersonalBusinessBusiness ProEnterprise
Price / month19169699Custom
Credits / month1,20010,00030,000Custom
Members138Unlimited
Agents (active)125Unlimited
Deployments (active)1 (internal chat)48Unlimited
Phone numbers12Custom
Knowledge-base pages8004,00012,000Unlimited
Upload storage (MB)2001,0002,500Unlimited

The Personal plan includes the internal AI chat workspace only — deployments to external channels (website widget, phone) require Business or higher.

See Team & Billing for plan changes, credit blocks, and auto-recharge.

What costs credits

Everything an agent does is metered in credits. Plan credits are spent first, then purchased credits (earliest-expiring first).

ActivityCost
AI model usage (text)Per ~1,000 words, in and out. The rate depends on the model (shown in the model picker).
Images in chatBilled as input text at the model's rate: roughly 1,000 to 2,000 tokens per screenshot, on every turn where the image is still shown to the model (the last three messages, or when the agent looks at it again).
Voice / phoneBilled separately and higher than text: per second of call plus audio tokens in/out.
Website crawling1 credit per page (a "page" ≈ 2,000 characters of extracted text).
Embedding (indexing)0.01 credits per 1,000 tokens.
Web search0.10 credits per search.
Web page read1 credit each time the agent opens a page in full.
Scheduled task runThe same as running that conversation by hand.

Note

A "page" is a unit of 2,000 characters of extracted text — used both for crawl billing and for your plan's knowledge-base page limit.

Supported file types and upload limits

TypeExtensions
PDF.pdf
Word.doc, .docx
Text.txt
Markdown.md
  • Maximum 100 MB per file. Larger files are rejected before upload.
  • Total upload storage is capped by your plan's Upload storage (MB) limit.

Chat attachments accept a wider set and a smaller size, because they are one-off documents rather than permanent agent knowledge:

Knowledge base sourceChat attachmentChat image
Types.pdf .doc .docx .txt .md.pdf .docx .txt .md .html .json .csv .xlsx.png .jpg .webp .gif
Max size100 MB25 MB10 MB, 10 per message, 25 per conversation
RetentionUntil you delete itOrganization-wide setting, 1 week to foreverSame as chat attachments

All three count towards the same Upload storage (MB) limit. Images are stored downscaled (long edge 1568 px) with camera metadata removed. See Attaching documents and Attaching images.

  • A scanned or image-only PDF with no extractable text is accepted but shows as Skipped — it contributes no searchable content. Use a text-based PDF or paste the text as Markdown. See Knowledge Base.

Channels

ChannelStatus
Website chat widgetLive
Phone / voice agentLive
Chat (Internal) — teamLive
WhatsAppComing soon
TelegramComing soon
EmailComing soon

Full details in Deployments.

Data-residency tiers

Set per organization under My Organization → General → Server Location. The tier filters which AI models you can select. Switzerland counts as European but is not part of the EU.

TierModels allowed
Swiss / SwissSwiss-hosted models from Swiss companies
Swiss / anySwiss-hosted models, any provider
Europe / EuropeEuropean-hosted models from European companies
Europe / anyEuropean-hosted models, any provider
GlobalNo restriction (default)

Platform infrastructure, knowledge-base storage, and chat history are in Switzerland (Infomaniak) regardless of tier. Voice telephony uses a US provider (Twilio). botts.ai does not train models on your data. See the FAQ and data-residency post.

Website widget embed

Each website deployment exposes a single script tag, pasted just before the closing </body> tag:

<script async src="https://YOUR-API-URL/widget/chat-widget.js?key=YOUR_API_KEY"></script>
  • The key is the deployment's API key, prefixed bts_. The real URL and key are shown in the widget editor's Embed Code section — copy them from there.
  • Regenerating the key invalidates the old snippet everywhere it's embedded.
  • Works on WordPress, Shopify, Squarespace, Wix, or plain HTML. See Deployments.

Connections and secrets

Connections let an agent call external systems. Configured under Integrations (Admin/Owner).

Connection typeWhere its tools come from
MCP serverDiscovered from the remote server, imported by you
HTTP APIEndpoints you define under one base URL, or import from an OpenAPI spec URL (JSON or YAML)
FieldValues
MethodGET, POST, PUT, DELETE, PATCH
AuthenticationNone, Bearer, Basic, Custom Header, or Login Token
Auth scopeOrganization-wide, or per user
Parameter typesstring, number, integer, boolean
Parameter optionsRequired, fixed value, restricted to allowed values
Live timeout30 seconds
Pipeline stepsunwrap, project, search, sort, top, render
  • Reference a stored secret anywhere in the URL, headers, or auth with $secret:key_name.
  • GET/DELETE send parameters as query string; POST/PUT/PATCH send a JSON body. Path variables like {order_id} are percent-encoded and substituted.
  • A 2xx response body is passed to the model, capped at both ends if very large; 4xx/5xx bodies are hidden (the model is told the status plus a hint not to invent data). Return 200 with a JSON body for "not found"-style answers.
  • Private, localhost, link-local, and cloud-metadata addresses are blocked, for MCP server URLs too.
  • MCP tools are registered as servername__toolname. A server that changes its tool list needs re-approval before the change takes effect.

Scheduled tasks

Personal recurring agent runs, under Scheduled in the sidebar. See Scheduled Tasks.

OwnershipThe account that created it. Not visible to other members, including Admins.
Who can createAny role that can use chat, including Member
RunsA deployed agent you can chat with, or a general-purpose model
FrequencyDaily, weekly (pick weekdays), or monthly (pick days 1–31), at a time in your timezone
DeliveryEmail, or in botts.ai marked unread. Saved to a conversation either way.
Auto-pauseAfter 3 consecutive failures, with an email
No creditsThe run is skipped, not queued or retried

Form fields

Built-in lead-capture templates (enable per agent under Form / Lead Capture):

TemplateFields (* = required)
Contactname*, email*, phone, message
Support Ticketsubject*, description*, email*, priority (low/medium/high)
Booking Requestname*, email*, date*, service

Custom fields use the same parameter types as tools — string, number, integer, boolean. One submission is captured per conversation; name/email/phone also fill the contact's empty profile fields. Submissions are viewable and exportable to CSV under each agent's Forms tab. See Agents.

What botts.ai doesn't do (yet)

So you (and any agent trained on these docs) don't assume a capability that isn't there:

  • No third-party knowledge-base connectors. Knowledge-base sources are website crawl and file upload only. There is no Google Drive, Notion, Dropbox, SharePoint, or S3 sync. (An MCP server can let an agent query such a system live, which is a different thing from indexing it.)
  • No public/customer API. There are no personal API keys or a general REST API for sending messages programmatically. The only keys are the per-deployment widget keys (bts_), which authenticate the embedded chat widget on public channels.
  • No calendar or booking automation. The Booking Request form collects a requested date; it does not check availability or book anything. Use a connection to call your own scheduling API.
  • No per-agent language selector. Agents reply in the language the user writes in; to pin a language, state it in the system prompt.
  • No OAuth for connections yet. Connections authenticate with keys, headers, basic auth, or a login endpoint. Catalogue entries whose only login is OAuth are listed but disabled.
  • No interactive voice failover. Phone deployments have no backup model.
  • WhatsApp, Telegram, and Email channels are not available yet (Coming soon).
  • No SSO/SAML or audit logs at this time.
  • No automatic prompt optimization. Testing grades an agent's replies against your expectations; it does not rewrite the prompt, pick a model, or change the agent in any way.

Need one of these? It's often achievable today with a connection, or worth raising with support.

Things this list used to say

Removed because they now exist, in case you remember them as limits:

Former limitNow
No web browsingAgents can search the web and read pages, optionally restricted to chosen domains
No conversation exportExport in JSON, CSV, or Excel, per message or per conversation
No document uploads in chatAttach documents to a conversation and ask about them
No scheduled or unattended runsScheduled tasks run an agent on a timer
Tools are hand-written HTTP endpoints onlyMCP servers and OpenAPI import

Last updated on September 13, 2026