AnAIagentthathaslearnedyourknowledgebase
Answers customers 24/7 in your Telegram bot. Built in an evening
The AI Map support agent on Telegram
Like a support rep — one that never gets tired
Answers on its own
From your knowledge base: prices, terms, instructions. At night and on weekends too
Calls a human
A question outside the base — a Telegram notification to the manager, the dialogue waits in the queue
Under your control
The whole dialogue, a single question or a pause — switched right from the queue
Built in an evening
An instruction, files, a bot from BotFather. Or let your Claude build it over MCP
evening to build — by hand or through Claude
models in a chain in case one is unavailable
free requests a day; with OpenRouter credits — 1000
we build and connect it — the team in Yerevan
Six agents from the same parts
One builder — different roles: support, sales, consulting, booking, a shop, comments
Support 24/7
The agent closes the frequent questions, at night and on weekends too. The hard ones go to a human with context. Tools: knowledge base, call a human
Sales and qualification
Finds out the task, budget and timeline, answers about the product, hands a warm lead to the manager with a CRM card. Tools: variables, CRM, webhook
Consulting on documents
Searches regulations and instructions, answers with the document as its source, calls an expert on anything non-standard. Tools: doc files, search
Booking and appointments
Offers free calendar slots, books, reminds; rescheduling — in the same dialogue. Tools: calendar slots, message
Shop bot
"Is this size in stock", "how much", "how is it delivered" — from the price list; the order goes to the manager, the buyer into the CRM. Tools: knowledge base, webhook
Replies in comments on call
A built agent can be called onto a specific comment or message in any channel right from the queue
Who it is, what it wants and where the limits are
The instruction is text: role, goal step by step, when to call a human. The model comes from a list: free ones for testing, Claude, GPT or Gemini for real work, up to three in a chain in case the first is unavailable.
- The /start greeting and reply rules — in the same card
- The brain is OpenRouter: you choose the model and the budget
- The agent's rights are dialogues only, like an operator's
Check it before customers see it
You write as a customer — and see the reply and the tool calls: which file the agent read, which action it fired. The same engine as in production, just without sending anything to customers.
- You can see where the answer came from — which knowledge-base file
- A mistake in the instruction is found in a minute, not after a customer complaint
- Test dialogues never reach the shared queue
The files the agent answers from
A faq file — one question, one answer, taken whole. A doc file — a large document: a price list, a regulation, a manual — the agent searches it for the right fragment.
- Upload in the agent's card — or over MCP from your Claude
- Updated the price list — the agent answers with the new one right away
- Each agent has its own base: support and sales don't mix up facts
Tell your AI which agent you need
Connect Chats to your Claude or ChatGPT with one command. The assistant writes the instruction itself, lays out the knowledge base, describes the actions and runs the tests — by our building rules.
- Our own example: the AI Map support agent answers from 25 documentation files and, when there's no answer, hands the question to the team. Built over MCP in one evening
- Don't want to do it yourself — the team in Yerevan will build it
claude mcp add --transport http aimap-chats https://chats.aimap.agency/mcp
Six parts — all in one card
Instruction
Role, goal step by step, limits, the /start greeting. The more specific the "don'ts", the fewer made-up answers
Knowledge base
faq — one question, one file, taken whole. doc — a large document the agent searches for fragments
Actions
Functions for the model and the steps behind them: a webhook, a Telegram notification, a message to the customer, dialogue variables, calendar slots
Channel
Your own Telegram bot from BotFather — there the agent works autonomously. In other channels — on call from the queue
Model
Free models to start without a budget; your own AI — Claude, GPT, Gemini — if you wish. Up to three in a chain
Test chat and built-in tools
The same engine without sending to customers; knowledge search, variables, calling the manager, tags and stopping the dialogue — every agent has them
See it without signing up
The Agents section of the demo workspace: instruction, knowledge base, actions and test chat
A demo workspace on fictional data; it opens on the official site, aimap.agency
An Armenian service. The team is in Yerevan
The agent builder is built, supported and rolled out by the AI Map team in Yerevan. For a company in Armenia that means a contract with a local legal entity, invoices in drams and support during Yerevan business hours.
A contract with an Armenian legal entity
Sole proprietor Leon Nikolaev, TIN 28154419, Yerevan. Contract, invoices and acts under the law of the Republic of Armenia; payment in drams to an Armenian bank account.
The implementation team lives and works in Yerevan
The founders and the AI Map team are in Armenia, not outsourced abroad. We implement online, in your time zone.
Support on Yerevan time
Questions get solved during your working hours, not when a team in another time zone wakes up. A direct line to the founder on Telegram.
The developer, not a reseller
The service's code is written by the AI Map team. Customisation for your process and integrations with your systems are done by the same people who built the product.
Russian now, Armenian on the roadmap
The service currently runs in Russian. Armenian, English and Korean versions are planned.
Already working in Armenia
The booking bot for Fabula Coworking, the kokasoku.com portal for the Eurasian Riichi Mahjong Association and Yerevan's Ere1 So club — all projects by the team.
CasesWrite, don't call: we reply on Telegram or WhatsApp during Yerevan business hours.
Three steps to the first reply — or one message to us
Build the base and the instruction
Articles and files — what the agent knows. The instruction — who it is, what it wants, where the limits are, when to call a human
Check it in the test chat
Write as a customer — see the reply and the tool calls: which file it read, which action it fired
Paste the bot token
A bot from BotFather, the manager's chat id — and the agent answers customers. Dialogues land in the shared Chats queue
Or call the team
We build the agent from your documents and correspondence, test it on real questions, connect the bot and train the managers. All online
Three ways to use it
Agents are part of the Pro plan together with chats, CRM and MCP. Turnkey building is a team service
Free
to get acquainted
- One project and three channels
- Contacts, tagging and analytics
- Control from your Claude or ChatGPT over MCP
- Chats, CRM and agents — on Pro
Pro
will become paid — users vote on the price
- Unlimited projects and channels
- Chats: queue, statuses, owners, templates
- CRM, contacts and automations
- AI agents and MCP control
- Support during Yerevan business hours
Enterprise
we run your chats or deploy on your side
- Channels, agents and CRM — set up turnkey
- 10 team hours a month (13 000 ֏/hour); 25 hours — from 260 000 ֏
- Infrastructure included, up to 3 000 AI requests a day
- Deployment inside your perimeter — on your servers
- Contract with an Armenian legal entity, invoices in drams
- A direct line to the AI Map team in Yerevan
Prices are in drams. Contract and invoices from sole proprietor Leon Nikolaev (Yerevan) under Armenian law. Sign-up needs no card.
Other services
Planner, chats, CRM and agents — one subscription, one database and one MCP
Frequently asked questions
Official service page — aimap.agency/products/chats/agents · Demo — aimap.agency/demo/agents · Documentation — aimap.agency/products/docs/agents
Build your agent this evening
Sign up without a card, free models for testing. Don't want to figure it out — the team in Yerevan builds the agent from your documents and connects the bot