Stop losing requests
The employee answers in your channels from approved materials and keeps the conversation moving when your team is offline.
Typical industry scenarios — no real client names. The figures are model estimates: open the calculator and run your own numbers.
Proof before promises
The bar below the number is the annual subscription price versus savings for the same year.
Model estimates on transparent assumptions — see the calculator.
From the first request to completed work
Do not deploy AI for its own sake. Pick one route, connect a source of truth, verify it on your data — then expand.
The employee answers in your channels from approved materials and keeps the conversation moving when your team is offline.
It reads orders, CRM records and documents through published plugins, then performs only the actions you allow.
Ask for a summary, exceptions or the next task in a normal conversation instead of collecting status by hand.
Two operators are swamped at peak and offline at night — order status, delivery and returns pile up.
An AI employee handles routine requests in Telegram and MAX around the clock (65% in this model scenario), passing only non-standard requests to a human.
Calls and messages get lost, bookings slip, patients leave for competitors.
An AI receptionist handles booking, rescheduling, reminders and visit prep in messengers 24/7.
Site, Avito and Telegram leads wait until morning — a notable share arrives off-hours.
An AI employee qualifies a lead in a minute at any hour, books a viewing and hands the manager a warm contact.
Tutors spend the day on routine questions about the syllabus, access and instalments.
An AI curator handles most routine questions (60% in this model scenario) and helps sell courses, freeing tutors to teach.
IT and HR specialists drown in repetitive tickets: access, how-tos, statuses.
A first-line AI employee in Bitrix24 answers staff instantly and files tickets itself.
Sales managers spend the day on repetitive queries — stock, prices, minimum order, shipment status — while big deals wait in line.
A sales AI employee answers stock and pricing and issues invoices in Telegram, WhatsApp and email around the clock, handing managers only large or non-standard deals.
Listing calls and messages arrive at night and in busy hours — some of them get missed.
A personal AI assistant answers on Avito and Telegram, qualifies and books viewings to the calendar.
Requests come in the evening, some are lost, and messaging eats teaching time.
An AI assistant takes requests, manages the schedule, reminds students and answers routine questions.
“Still available?” arrives any time; a slow reply loses the buyer to the next seller and drops your ranking.
An AI employee replies in seconds any time, closes the deal and passes non-standard questions to a human.
Below is what is already assembled and running. The comparison with frameworks and coding agents opens next, with links to the sources.
Self-hosted, conversations, credentials and knowledge live on your infrastructure.
Works on free and local models, with optional BYOK.
Channels, long-term memory, RAG, voice, marketplace and an admin UI in one platform.
One light server plus PostgreSQL, installed with a single command.
Separate workspaces, SSO (SAML/LDAP/SCIM), Row-Level Security, approval of risky actions and air-gapped mode on Enterprise; RBAC, audit and White-label from Business.
MAX and Bitrix24, SIP telephony, a focus on Russian infrastructure; the Russian software-registry application has been filed.
Cloud bots charge per dialog, per resolution or per token — your bill grows with your traffic. AiHummer is a fixed platform-licence price: growing load does not change it, though model and infrastructure costs may grow.
One product, two deployment models
AiHummer runs the infrastructure. You set up the employee, knowledge and channels in the portal and can begin with a free trial.
Start freeConversations, credentials and knowledge stay on your host. Use the documented installer and keep control of the environment.
Open quick startChoose which providers, channels and plugins the employee may use. Self-hosting does not magically isolate external services: each enabled integration receives only the data required for its request.
Start in the cloud, or first challenge the public demo with the questions your customers actually ask.
Write to a working employee yourself. The stand is open to everyone and resets every hour.
Open the demo 02The number is read from the current catalog snapshot, not typed into this page.
Open the catalog 03Change workload and cost assumptions yourself. A model is not presented as a customer result.
Calculate your caseServices in this ribbon: 115. Which of them are open depends on your plan. View all plans →

Your data
Your server or your cloud
What crosses the line
Free or your own models
To the model — only the question and the piece of document it needs
And only if the model is external. A local one works with no internet at all.
AiHummer
To us — a licence check
Install id, plan and version. We never receive your conversations, memory or files.
152-FZ: since 01.07.2025 primary PII collection must be in Russia; repeat-breach turnover fines of 1–3% of revenue, up to ₽500M.
Deployment
Cloud
Personal Helper
Free
$16/mo after 10 days
$13/mo after 10 days when paid yearly
instead of $192 month by month
You save $32 a year
Russian data centre, we run it
Get startedSelf-hosted
Community
Free
Free, with no time limit
Self-hosted on your hardware.
Starter$69$58 when paid yearly
Free or your own models
Open quick startChange workload and cost assumptions yourself. A model is not presented as a customer result.
Get startedIt is a calculation on stated assumptions, not a measurement at a customer. Salaries come from a market reference, payroll taxes are 30.2%, and the requests-per-person norm and routine share are named under “How this is calculated”. The “Calculate for yourself” button opens the calculator pre-filled with that case's numbers.
No, and we do not hide it: there is not a single figure about somebody else's results on this site. The cases are typical workloads costed with our model. You can check the product by hand — the demo stand is open and resets once an hour.
Those are libraries: you assemble an agent from them and write everything around it yourself — channels, memory, permissions, a control panel, billing, updates. AiHummer is a finished program where that already exists. The comparison table across twelve capabilities and twenty-two systems opens above.
On what happens as you grow. Cloud bots charge per dialogue or per resolved request, so the bill grows with traffic; the AiHummer license does not depend on load. The second tab of the calculator works out the monthly volume from which the license is cheaper — put in your own per-dialogue price.
Installing is a single command from your account area. After that the time goes into your data: connect a channel, upload documents, describe what the agent does on its own and where it must hand the conversation to a person. Implementation and training can be ordered as a separate service.
It hands the conversation to a person. What the agent decides on its own and what it must pass on is set by a permission boundary in the control panel — a setting, not a guess by the model.
Telegram, MAX, VK, WhatsApp, Avito, e-mail, Bitrix24 and SIP telephony. Which channels are open on which plan is shown on the plan cards; on the free Community plan only Telegram is.
Yes. Below the business cases there are personal ones — an assistant for family, study, health and the home. The cloud has separate plans for them, and on your own server the free Community plan is enough.
Yes. The demo stand is open around the clock and resets once an hour, so you can poke at it as long as you like without breaking anything. The memory section and the module catalogue are there too.
Install AiHummer on your own server, or take a ready one in the cloud — the Community plan is free.