01
The assistant on every screen
Piper opens beside any page and knows the page and the selection, so “why is this one stuck?” needs no explaining. One box takes every request — answer, investigate or build.
Conversational AI
Conversational AI that knows your work.
Not a chatbot in a corner. Agents people talk to wherever they work — answering from your data, citing it, and able to act.
On Botminds every agent holds a conversation: beside every screen, on the home page, inside apps, by email, in Microsoft Teams and through an API — with Viber and Telegram bringing customer messages in. Built for a project on knowledge bases of millions of documents, it is the same agent at every door: the same knowledge, tools and record.
On every screenthe platform assistant, beside your work
Home pageyour own agents, a click away
An agent’s workspaceits chat beside its data and apps
Inside appsa chat card on any page
One agentsame knowledge · same tools · same rules
Emailthe agent’s own address
Microsoft Teams@mention it, or chat one to one
Viber and Telegramcustomer messages in, cases opened
Your systemsan OpenAI-compatible API
In the appWhere people already are
What is conversational AI?
Conversational AI is software people talk to in their own words — by chat, email or voice — that understands the request, keeps track of the conversation and replies in natural language. Enterprise conversational AI adds what a business needs: answers drawn from the organisation’s own data, held to what each person may see, with sources shown and actions a person approves.
Scales out on KubernetesServices and workers add capacity as work arrives, and give it back when it is done.
Built for millionsMillions of documents, records and agent runs — the indexes and workers grow with them.
Any cloud, any regionAzure, AWS, Google Cloud or on-premises, in the region your data must stay in.
Always onRedundant services, rolling releases and continuous backups, for mission-critical work.
Secure and governedSingle sign-on, encryption, least privilege and a sealed audit trail.
Conversational AI vs chatbot vs agentic AI
Conversational AI vs chatbot vs agentic AI.
The difference that matters to an enterprise is not how fluent the replies are — it is where the answers come from, who they reach, and whether the system can act on what it was told.
01Scripted chatbota decision tree with a chat window
- Understands
- the phrases someone wrote down in advance
- Answers from
- replies written in advance
- Can do
- hand over a form or a link
- When unsure
- “Sorry, I didn’t get that”
02Conversational AIa language model that talks
- Understands
- whatever people type, in their own words
- Answers from
- your documents and data — or its own memory
- Can do
- explain, summarise, draft
- When unsure
- may sound sure anyway
03Agentic conversational AIa conversation on top of an agent
- Understands
- the words, the person, the page they are on
- Answers from
- the sources it was given, within what the asker may read — cited
- Can do
- act through granted tools — with approval on the tools you choose
- When unsure
- searches again, or says the sources do not cover it
Inside the platform
Conversations wherever people already work.
On Botminds the conversation sits where the work is — beside the screen, on the home page, inside the app built for the job.
02
Personal agents on the home page
Each person keeps their own agents, a click from the home page, with the same composer and features as the assistant.
03
An agent’s own workspace
The chat, and beside it the collections, pages and apps the agent works with — shared with everyone the agent is shared with.
04
Chat cards inside apps
Any page of an app carries a chat card bound to one agent: suggested questions, streamed answers, sources beside the chat, earlier conversations.
05
Shared conversations
Share a conversation with colleagues to watch it or to continue it together, signed in with their own accounts.
06
Approvals in the thread
You choose which tools wait for a person. When the agent reaches one, the request appears in the conversation: approve, reject or edit.
Conversational AI at enterprise scale
Thousands of conversations at once, every channel, the same agents.
The chat runtime holds no conversation in memory: conversations, summaries and files are stored durably, so replicas scale out as people arrive and any replica takes the next turn. Customer messages queue durably and workers scale with them — on Azure, AWS, Google Cloud or on-premises.
Live conversations
Every screenHome pageAppsEmailMicrosoft TeamsAPI
Customer messages
ViberTelegramMailboxes
Chat runtime · stateless replicas
scale out with the conversations · answers stream back as they are written
Durable queues · workers
bursts absorbed · workers scale with the queue · every message kept until worked
The same agentssame knowledge · tools · rules · record
Stored durably
conversations, summaries and files — any replica takes the next turn
What happens in one turn
One conversational turn, from message to record.
Behind a reply that takes a few seconds, the same steps run every time, at every door.
- Message
Typed, dictated, or with files attached.
- Context
Who is asking, the page they are on, the conversation so far.
- Search
The agent’s sources, within what this person may read.
- Act
Granted tools; the ones you chose wait for a person.
- Answer
In the person’s language, with its sources — kept with the conversation.
How the context is assembled: context engineering. How the search decides where to look: agentic RAG. How actions are granted and approved: the agent harness.
Chat with your data
Chat with your data — every answer cited.
The risk with enterprise conversational AI is a confident answer drawn from a document the asker should never see, or from nothing at all. Both are closed off by design — across knowledge bases of millions of documents.
A conversation grounded in a firm’s own records
You
How many supplier invoices are still unmatched this quarter, and which is the largest?
Agent
42 invoices are unmatched for Q3. The largest is Bayside Freight’s INV-88213 for $48,920 — its purchase order was closed early. Sources: Invoices table · PO-5521
You
Draft a note to Bayside asking for a corrected invoice.
Agent
Here is the draft. Sending it uses the email tool, which waits for your approval — approve, reject or edit it.
- Same rules as the screensIn the app, a collection’s read rules apply to the agent’s search and its queries: what a person cannot open, their conversation cannot quote.
- Counted, not guessed“How many” and “how much” go to the records as queries.
- Sources shownThe documents and pages the answer used, opened beside the chat.
- Knowledge lockChoose to have an agent answer from its sources alone.
- Reachable agentsBy email, in Teams or over an API key, an agent searches what it was given — you choose what it holds.
Attachments and voice dictation
Bring a file into the conversation, or just say it.
People rarely ask in a vacuum: they have the spreadsheet, the photo, the contract. They drop it into the chat and ask about it — or dictate the question.
Q3-supplier-invoices.xlsxspreadsheet · readdamaged-pallet.jpgphoto · read
Which of these invoices don’t match the pallet in the photo
Listening…Send
What happens to a file
- Readtext, tables and images, page by page
- Indexedlike any library document
- Searchedinside its own conversation
- Citedwith the page it came from
PDFWordExcelPowerPointtextMarkdownCSVJSONPNGJPEGWebP
Kept to its conversation — for its owner and everyone it is shared with.
Kept to its conversation
A dropped file is searched inside its own conversation, for its owner and everyone it is shared with.
Indexed like any document
The same reading and indexing as a library document, so the agent finds page 97 and quotes it with its page.
Voice in the browser
Dictation runs on the browser’s own speech recognition, and a one-time notice tells people who processes the audio.
Agent Reach
AI agents in Microsoft Teams and email — with their own address.
Agent Reach takes the agent to people who will never open another app: a mailbox of its own and a presence in Microsoft Teams, answering there as itself — not as a person, and not from a person’s inbox.
Email
- A mailbox of its ownan address created for it — never a person’s, never an existing team mailbox
- Your colleaguesanswers people writing from inside your domain
- One thread, one conversationa reply continues where the thread left off
- Never answered twicereplies come from the agent’s own mailbox
Microsoft Teams
- Installed onceone app for all your agents, added by a Microsoft 365 admin
- @mention or one to onename the agent in a channel, or pick it in a private chat
- Sealed to your tenantmessages from any other organisation are refused
- Answers in the threadwith “switch agent” to talk to another
Same engine as the API— so pausing the agent, spend quotas and the run record apply to every email and every Teams message.
Promoted by an admin
An admin promotes an agent to email, Teams or both; one Teams app serves every agent in the organisation.
Paused in one switch
An operator pauses a promoted agent in one move; its mail waits, unread, for the resume.
On the record
Every email and Teams turn is kept on the run record like a chat in the app, under the same spend quotas.
Messaging channels
Viber and Telegram: customer messages in, cases opened.
Customers write on the messaging apps they already use. Connect Viber and Telegram as channels and every message, with its photos and files, becomes a record the right agents and people work on.
- Connected
A Viber or Telegram account is connected to a collection, like any source.
- Received
Messages and their media land in the collection as records.
- Opened
Switch on cases and a new message opens one, with its own reference.
- Acknowledged
The sender gets a confirmation with that reference.
- Worked
The collection’s agents, workflows and people take it from there.
For your own apps
An OpenAI-compatible API for every agent.
Any tool that talks to a chat-completions endpoint can talk to a Botminds agent: point it at the platform, name the agent as the model, and use a key. The agent keeps its own instructions, tools and knowledge.
POST /api/runtime/v1/chat/completions
Authorization: Bearer bm_live_…
{
"model": "agent:supplier-desk",
"stream": true,
"messages": [
{ "role": "user",
"content": "Which invoices are unmatched?" }
]
}- Two shapesChat completions for any compatible client; responses with stored conversations, background runs and cancel.
- The agent owns its personaIts instructions and tools stay its own, whatever a client sends.
- StreamingAnswers stream as server-sent events.
- Safe retriesSend an idempotency key on the responses endpoint and a retried request reuses the first.
- KeysProject keys stored hashed, or OAuth client credentials.
Other languages
Answers in your language — screens translated in your browser.
The agent replies in the language it is spoken to in. On app screens, the words that come from your own data are translated on the reader’s own device where the browser supports it, and otherwise by the browser’s own translator. Botminds runs no translation server.
The conversation
Quels contrats fournisseurs expirent avant décembre ?
Trois contrats expirent avant décembre : Northwind, Bayside et Arden…
The agent replies in the language of the person’s newest message.
The screen
Does this browser have a built-in translator?
yesTranslated on the devicenames, labels, stages, values, headings
noLeft as writtenwith a pointer to the browser’s own translate button
Never translated:numbersdatesamountsemail addresseslinksIDsfile namescodedocument text
0 characters sent by Botminds to a translation server. A browser’s own translate button uses that browser’s service.
Side by side
Choosing a conversational AI platform for the enterprise.
Most enterprise chatbot projects stall on the same questions: whose data the bot can see, what it can do, where people reach it, and whether it holds up at scale.
| Chatbot builder | General AI assistant | Botminds conversational AI | |
|---|---|---|---|
| Answers from | Scripted replies and an FAQ | Whatever the user pastes or connects | The libraries, documents and records the agent was given — millions of documents |
| Who sees what | Everyone sees the same | Depends on each connector | Each person’s read rules, inside the search |
| Can act | Hand-built integrations | Some, with plug-ins | Granted tools, with approvals in the thread on the tools you choose |
| Sources | None | Sometimes | The sources the answer used, to the page |
| Where people reach it | A website widget | Its own app | Every screen, home page, apps, email, Microsoft Teams, Viber and Telegram intake, API |
| Scale | Per bot | Per seat | Stateless replicas that scale out; durable queues; any cloud or region |
| Record | Chat logs | Chat history | Every turn, tool call and approval kept with the run |
Questions people ask
Conversational AI, answered.
More from the platform: the full architecture, and how security works at Botminds.
What is conversational AI in simple words?
Conversational AI is software you talk to in your own words and that replies in natural language, keeping track of the conversation. In an enterprise it should also answer from the company’s own data, respect who may see what, and show where each answer came from.
What is the difference between conversational AI and a chatbot?
A traditional chatbot follows a script: it recognises phrases someone wrote in advance and returns prepared replies. Conversational AI understands free text and composes its replies. On Botminds the conversation also sits on a full agent, so it can search, cite and act.
Conversational AI vs agentic AI — what is the difference?
Conversational AI is about the dialogue: understanding and replying. Agentic AI is about doing: planning steps, using tools and finishing tasks. Botminds combines them — every conversation runs on an agent with granted tools and the approvals you choose. See the agent harness.
Can conversational AI handle enterprise scale?
Yes. The chat runtime holds no conversation in memory, so replicas scale out on Kubernetes as conversations grow and any replica takes the next turn. Customer messages queue durably and workers scale with the queue, on Azure, AWS, Google Cloud or on-premises.
Can I chat with my company’s data safely?
Yes. In the app, the agent searches the sources it was given, within what the person asking may read under each collection’s read rules, and shows its sources. Agents reached by email, in Teams or over an API key search what they were given, so you choose what each holds. See agentic RAG.
Can AI agents work in Microsoft Teams?
Yes. With Agent Reach, one app installed once by a Microsoft 365 admin lets people mention an agent in a channel or pick it in a one-to-one chat. Messages from other organisations are refused.
Can an AI agent have its own email address?
Yes. An admin gives an agent a mailbox of its own — an address created for it, never a person’s. It answers colleagues inside your domain, one thread as one conversation, and answers each message once.
Which messaging channels feed Botminds?
Viber and Telegram connect as channels: customer messages and media become records, open a case with its own reference and receive a confirmation, and the collection’s agents and people work them. Conversations with agents also run in the platform, by email and in Microsoft Teams.
Is there an API for Botminds agents?
Yes. Every agent is reachable through OpenAI-compatible chat-completions and responses endpoints, with streaming, idempotency keys on the responses endpoint, and project API keys or OAuth client credentials.
Does it work in languages other than English?
The agent replies in the language of the person’s message. On app screens, text that comes from your data is translated on the reader’s own device where the browser supports it, and otherwise by the browser’s own translator. Botminds runs no translation server.
Learn the basics
AI agents in Microsoft TeamsHow agents reach Teams, the ways to build them, and how admins govern them→Conversational AI vs agentic AISystems that talk, systems that act, and the ones that do both→Conversational AI vs chatbotScripted replies, matched intents, and models that understand free text→All guidesEvery explainer, by topic→
An agent your team can just ask.
Bring the questions your team asks every day. We’ll show you an agent answering them from your data — in the app, by email and in Teams.