Answers from your documents
Catalogues, instructions, policies, procedures and FAQs become usable conversation context.
← All projects
![MIRA Knowledge Telegram screen with an answer from company knowledge]()
AI assistants · knowledge base · Telegram
MIRA Knowledge — the AI assistant that knows your company
The company uploads its documents and the client asks questions in ordinary Telegram messages. The bot finds relevant information, uses previous messages and replies in the company style — without manual file searches or repetitive manager answers.
Working Telegram service·RAG, LLM, webhook, admin control
Documents → context → Telegram answer
01Your documentsThe company defines what the bot can answer
02A natural questionThe client writes naturally instead of searching a menu
03A clear answerFacts come from the knowledge base; complex cases go to a manager
Business value
Company knowledge works in every conversation
The bot removes repetitive questions from the team without becoming a rigid autoresponder: it uses internal information, remembers the conversation and knows when a human is needed.
The client writes naturally
There is no need to guess a menu button: the bot understands questions, follow-ups and topic changes.
Remembers recent messages
Questions like “what about this option?” or “how long will it take?” do not require repeating the whole context.
The manager sees the conversation
The administrator sees who asked, what they asked and what answer the bot sent.
The knowledge base can be updated
After documents are added or changed, the administrator reindexes the base without changing the interface.
Errors do not break the service
If the model, knowledge base or a component fails, the client gets a clear fallback and a way to reach a manager.
How it works
Not just an AI chat — a business workflow
The model is not simply told to answer. The system finds relevant company fragments and passes them with the conversation rules before generating a reply.
Company documents
MD, TXT, PDF and DOCX: catalogues, instructions, policies and frequently asked questions.
Relevant context
A local index finds the fragments related to the current client question.
The LLM as MIRA
The model receives rules, context and recent messages to reply naturally and stay on point.
Telegram and the manager
The client gets an answer while the team sees complex questions and can join manually.
How it looks
A dialogue that does not feel automated
From the first message to a complex follow-up, the interface stays familiar and the replies feel like a real consultation.
1. Clear beginningThe client immediately understands what to ask and what the bot can do
2. Answer from company knowledgeThe model uses internal documents instead of inventing terms
3. Conversation continuityRecent messages make follow-ups possible without repetition
4. Administrator controlThe team sees the client question and the answer the bot gave
Workflow
From a file in a folder to a ready consultation
01 / KNOWLEDGE
The company adds documents
Materials go into one folder: instructions, catalogues, rules, FAQs and internal procedures.
02 / INDEX
The bot builds the knowledge base
Documents are split into chunks and become available for semantic search.
03 / DIALOGUE
The client asks a question
The client can ask about products, timing, rules or continue the previous topic.
04 / CONTROL
Complex cases go to a human
If the data is insufficient or an individual calculation is needed, the bot suggests a manager and notifies the team.
Where it fits
One mechanism — many business tasks
This setup fits companies that already have documents but still answer the same questions manually for clients and employees.
✓
Catalogue assistant
Explains products, specifications, materials, options and ordering conditions.
✓
Customer support
Answers about delivery, warranty, returns, payment and routine requests.
✓
Internal team assistant
Helps employees quickly find rules, instructions and work procedures.
✓
Lead qualification
Captures the essence of the request and gives the manager useful context instead of a vague “call the client back”.
✓
Multiple knowledge sources
Documents, FAQs, instructions and changing materials can be combined in one access point.
✓
Telegram without a separate account
The client does not need a new website account; they ask where they already communicate.
For the administrator
Control without unnecessary interface
The administrator does not need a complex dashboard. Telegram keeps only useful service actions: knowledge status, document updates and assistant notifications.
01
Update the knowledge base after changing files without restarting the bot.
02
Receive a client question together with the answer they received.
03
Preserve the history and keep the request even during a temporary model error.
Under the hood
Technology responsible for the actual result
The complex part stays inside. For the client it is a normal Telegram dialogue, while the system manages knowledge, context, limits and resilience separately.
AreaTechnologiesBusiness value
InterfaceTelegram, aiogramA familiar communication channel without a separate account or complex dashboard
Knowledge searchPython, TF-IDF, local indexThe bot finds relevant internal document fragments before answering
AnswersOpenAI-compatible LLMNatural dialogue, follow-ups, small talk and careful context handling
HistorySQLite, last 5 messagesConversation continuity without repeating the task
AdministrationReload, limits, notificationsKnowledge updates and request control without exposing service functions to clients
DeploymentDocker, Nginx, HTTPS webhookAn isolated service with automatic startup and Telegram webhook events through a domain
Need an assistant for your knowledge base?
Turn documents into a working service
I will build a Telegram bot that answers from your materials, reduces routine work and passes only important requests to a manager.