In brief: Alice AI for Business is Yandex’s enterprise AI assistant, launched on September 8, 2026. It works inside Yandex 360, can access Mail, Drive, Calendar, Tracker, and Wiki on its own, and through plugins connects to 1C, Bitrix24, amoCRM, and SPARK. In “Tasks” mode, it handles multi-step assignments: turns meeting takeaways into tasks, processes customer requests, and gathers data from invoices and contracts. For companies on Yandex 360, it’s the fastest way to try agentic automation: free for three months, no coding required. But this is a ready-made cloud service. If data can’t leave your environment, you need custom skills or a nonstandard integration, you’ll have to wait for the hybrid and on-premise version or build the solution yourself — for example, on the open model Alice AI Foundation under the Apache 2.0 license.
This material is for owners, CIOs, and department leaders deciding whether to adopt Alice AI for Business and what to do next. Facts were verified on October 3, 2026 against Yandex’s press release, the product page in Yandex 360, and the model card on Hugging Face. We did not conduct our own testing of the assistant. Pricing after the trial period and the full feature set may change — check the official page. AIrassvet is not affiliated with Yandex through partnership or advertising arrangements and does not receive compensation for mentioning products.
Contents
- What Alice AI for Business Is
- Three Products Under One Name
- What the Assistant Can Do in Company Departments
- Integrations: Yandex 360, 1C, Bitrix24, amoCRM
- Data and Security
- Alice AI Foundation: an Open Model for Your Own Services
- When the Ready-Made Assistant Is Enough and When You Need Your Own System
- How to Run a Pilot During the Three Free Months
- Common Implementation Mistakes
- Frequently Asked Questions
- Primary Sources and Verification Method
- How AIrassvet Helps Implement an AI Assistant in Company Workflows
- Bottom Line: Alice AI Is a Fast Start, but Not the Whole Architecture
What Alice AI for Business Is
Alice AI for Business is an enterprise AI assistant in the Yandex 360 ecosystem that not only answers questions but also executes chains of actions across a company’s work tools. Yandex introduced it on September 8, 2026, and described it as an assistant “for companies of any size,” from retail to construction (Yandex press release).
The assistant has two modes (product page):
- Chat — for quick tasks: rewrite an email, check a policy, summarize a long document, draft text, or generate an image.
- “Tasks” — for multi-step work: the assistant finds data in emails and documents on its own, processes it, and assembles a finished result.
It’s the “Tasks” mode that makes the product agentic. An AI agent is a system that receives a goal, chooses the steps and tools itself (email, CRM, files), and carries them out instead of just generating text. We covered the difference between an assistant and an agent in the article “Enterprise AI Assistant for Employees”.
At launch, the assistant was available to all Yandex 360 customers — according to Yandex, that’s more than 185,000 organizations. New companies can sign up for Yandex 360 and get a three-month trial, during which the assistant is free (press release). It’s available at alice.yandex.ru, in the mobile app, and in the Mail widget.
Three Products Under One Name
In September 2026, Yandex released several products under the Alice AI name, and they’re easy to mix up. The difference matters: it determines where data is stored and who is responsible for customization.
| Product | What it is | Where it runs | Who it’s for |
|---|---|---|---|
| Alice AI for Business | Ready-made assistant with the agentic “Tasks” mode | Yandex 360 cloud | Companies that already work in Yandex Mail, Drive, and Tracker |
| Customizable assistant version | Assistant with custom skills and plugins, hybrid or on-premise deployment | Yandex Cloud, hybrid, own environment (announced) | Companies with special data-storage requirements |
| Alice AI Foundation | Open 80B-A3B base language model under Apache 2.0 | Anywhere: on your own servers or in the cloud | Teams building their own AI service |
Yandex announced the customizable version in the same press release: it will be deployable in Yandex Cloud, in a hybrid setup, or in your own infrastructure, with the ability to create custom skills and plugins, plus advanced security and data-storage settings (press release). In business coverage, it’s called “Alice AI Pro” (T-Business Secrets). As of October 3, 2026, we did not find official sources with the timeline or commercial launch terms for this version.
What the Assistant Can Do in Company Departments
Yandex provides use cases for specific departments (product page). Below are the officially described ones, along with our view of where they deliver the most value.
| Department | Yandex-described scenario | What to pay attention to |
|---|---|---|
| Leadership | Preparing data for meetings, assigning tasks to owners | Tasks are created in Tracker; if your team works in Bitrix24, check the plugin |
| Sales | Finding deals that are stuck, reminders about customers not contacted in a while | Works only if the CRM is connected via plugin and its data is up to date |
| Marketing | Analyzing market and ad campaigns, turning findings into a presentation | The assistant may interpret external data inaccurately — verification is needed |
| HR | Matching resumes to role profiles, answering policy questions for new hires | Hiring decisions remain with people; account for personal data requirements |
| Support | Finding recurring issues in tickets, instructions, and knowledge base answers | A good first pilot: the impact is easy to measure |
| Finance | Summaries of data from different systems, planned vs. actual, anomalies | Numbers should be verified against the accounting system before decisions are made |
A knowledge base is worth calling out separately: you can collect policies and instructions there, and the assistant will answer based on them with source links. It's the same principle used in RAG systems: RAG (Retrieval-Augmented Generation) is an approach in which the model looks up relevant passages in your documents before answering and relies on them instead of its own memory. Source links let you verify an answer in seconds. We covered how these systems work and where their accuracy limits are in the article “RAG System for Business”.
The assistant can also upload contracts and spreadsheets, compare versions, find the needed numbers, and use them to assemble a new document — a report, mailing, or presentation text. For comparison: in the same month, Yandex updated its “NeuroLawyer” service, which compares the counterparty’s edits with the original contract from a single prompt and outputs a DOCX with corrections (T-Business Secrets). That does not replace a final review by a lawyer.
Integrations: Yandex 360, 1C, Bitrix24, amoCRM
Out of the box, the assistant works with Yandex 360 services — Mail, Disk, Calendar, Tracker, and Wiki — and corporate systems from other companies connect in the “Plugins” section: Bitrix24, SPARK, amoCRM, and 1C (product page).
What this means in practice:
- If your document workflows and tasks are already in Yandex 360, the assistant gets context right away: emails, files, meetings, tasks. That's the main advantage over generic chatbots.
- If most of your work runs in 1C or Bitrix24, it all comes down to how deep the plugin goes: which objects it can see (documents, directories, deals), and whether it can only read or also create records. We couldn't find public detailed documentation on the permissions for each plugin as of October 3, 2026 — check this during the pilot.
- If you use an industry-specific system or custom-built CRM, there is no ready-made plugin. The announced customizable version promises custom plugins, but until it launches you'll need to build the integration separately — for example, through an agent connected to your system's API. We described how this is done for 1C in the article “How to Connect AI Agents to 1C”.
A simple test that business publications also recommend: time how long it takes to assign tasks after a meeting manually, then compare it with the assistant's work (T-Business Secrets). That is your baseline for the pilot.
Data and security
On the product page, Yandex states four security measures (product page):
- Organization data is not used to train models.
- Data is stored in Yandex's Russian data centers.
- Corporate and personal profiles are separated, so work data is not mixed with personal data.
- The administrator manages access, and single sign-on (SSO) is supported.
For most small and midsize businesses, that's enough to start a pilot with non-confidential processes. But there are cases where a cloud assistant is not a fit, regardless of the vendor's claims:
- regulator or customer requirements prohibit sending certain data to any external service;
- you need a full audit trail of the assistant's actions in your logging system, not just in the vendor interface;
- you process large volumes of personal data and must control every processor.
For such cases, Yandex announced hybrid and on-premise deployment. An alternative available now is to deploy the language model within your own environment. We calculated the cost pros and cons of that path in the article “Cloud or On-Prem LLM: Total Cost of Ownership”.
Also keep regulation in mind. Since September 1, 2026, Law No. 243-FZ on supporting AI development has been in force; requirements for labeling generated content and informing users about rights to the output will take effect on March 1, 2027 (T-Business Secrets). If you're building a customer-facing service on AI, these points should be built into your user documents in advance. More details in the analysis “AI Law 243-FZ: What Businesses Need to Do”.
Alice AI Foundation: an open model for your own services
Alice AI Foundation (full name: AliceAI-Foundation-80B-A3B-Base) is Yandex's base language model, trained from scratch and published on Hugging Face under the Apache 2.0 license, which allows commercial use (model card). The model card was created on September 12, 2026, and Yandex published a detailed technical breakdown on Habr.
Key specs from the model card:
| Parameter | Value |
|---|---|
| Architecture | Hybrid, with Mixture of Experts (MoE) layers |
| Parameters | 80 billion total, 3 billion active per token |
| Context | up to 262,144 tokens |
| Layers | 48: 36 Kimi Delta Attention layers and 12 Gated Attention layers, each with MoE |
| MoE experts | 512, with 10 plus 1 shared selected per token |
| Training stage | pretraining (base model, not the chat version) |
| Languages | Russian and English |
MoE (Mixture of Experts) — an architecture in which a model is made up of many "experts," but only a small subset is activated for each token. So with 80 billion parameters, only about 3 billion are active per token: less compute, more knowledge. But the weights for all experts still have to stay in memory — for an on-premises server, that matters more than the number of active parameters.
What Yandex’s benchmarks show (5-shot, internal infrastructure, vLLM, temperature 0; all figures are from the model card:
| Benchmark | Alice AI Foundation | DeepSeek-V4-Flash-Base (284B-A13B) | Qwen3.5-35B-A3B-Base |
|---|---|---|---|
| WikiWebFacts (facts in Russian) | 86.5 | 83.2 | 62.4 |
| HardMultiQA (facts in Russian) | 67.9 | 65.4 | 47.2 |
| ExpertFactsQA Law (legal) | 49.6 | 40.5 | 27.9 |
| MMLU-Pro CoT (English) | 66.8 | 66.5 | 63.2 |
| TriviaQA (facts in English) | 79.0 | 89.4 | 71.4 |
| FinQA 128k (financial reports, long context) | 74.1 | 74.1 | 73.5 |
The takeaway from the table: the model is especially strong at factual knowledge in Russian, but it lags behind larger competitors on English-language facts. Keep in mind that these are the developer’s own measurements on proprietary benchmarks (Yandex released WikiWebFacts and HardMultiQA with the model); they do not replace independent testing on your own tasks.
The main limitation for business use: this is a base model. It can continue text, but it is not trained to follow instructions or hold a conversation. Business publications note directly that before launching your own service, you will need to adapt it (T-Business Secrets). That means:
- for a chatbot or agent, you will need instruction tuning and safety testing of the responses;
- an open-weights model can be deployed in your own environment, but you are responsible for the infrastructure, updates, and monitoring;
- if you need a ready-to-use chat model right now, compare alternatives — we covered them in the review "Local LLMs 2026: Models, Hardware, and Cost".
When a ready-made assistant is enough, and when you need your own system
We choose between three paths based on four questions: where the data lives, how standard the process is, whether data can be moved to the cloud, and who will support the solution. Below is our editorial decision framework.
| Situation | What to choose | Why |
|---|---|---|
| Email, files, and tasks are already in Yandex 360, and the processes are standard | Alice AI for Business | Minimal setup, free pilot, context from the services is already available |
| The main work is in 1C or Bitrix24, and you only need read access and summaries | Alice AI for Business with a plugin | Check in the pilot which objects the plugin can see |
| The assistant must create documents and deals in the accounting system according to company rules | Your own agent integrated with the system | You need precise permissions, checks, and an activity log |
| Data cannot be transferred to an external cloud | A customizable version (after launch) or your own LLM in your environment | A ready-made cloud service does not meet the requirements |
| You need your own AI product for customers | Your own system built on an open model | Control over the model, data, and user terms |
The tradeoff for choosing a ready-made assistant is dependence on the Yandex 360 ecosystem and on which plugins and permissions the provider opens up. The tradeoff for building your own system is development and support. That is why a sensible strategy for many companies is hybrid: give employees’ standard tasks to the assistant, and build critical processes that write to accounting systems separately and under full control. An exit plan also helps: decide in advance how you will move your knowledge base and workflows if you change providers ("Exit Plan from an AI Service").
How to run a pilot in three free months
A three-month trial is enough time to produce measurable results if you stay focused. Step-by-step plan:
- Pick one process with a clear outcome: handling customer inquiries, assigning tasks after standups, or summarizing overdue deals.
- Establish a baseline: how long the process takes now, how many errors or misses there are, and how many people are involved.
- Define the acceptance criterion: for example, "tasks after the meeting are created in 5 minutes instead of 30, with no missed assignments" — substitute your own numbers.
- Check the data: are the knowledge base guidelines current, are the CRM fields filled in? The assistant will not fix bad data.
- Set access permissions through an administrator: only the employees who need it and only the services they need.
- Connect the plugins to 1C or a CRM system and check what the assistant can see and change.
- Run it with 2–3 employees for 2–4 weeks and collect error cases.
- Compare it with the baseline and calculate the time saved in hours per month.
- Decidewhether to scale, limit it to specific tasks, or move to a custom integration.
More on how to choose your first process and avoid metric mistakes in these articles: “Implementing AI in Business Processes: Where to Start” and “AI Agent Performance Metrics”.
Common implementation mistakes
- Turn it on for everyone at once. Without a chosen process and metric, after three months it’s impossible to tell whether there was any impact.
- Mixing up products. The open-source Alice AI Foundation is not a “free Alice for business,” but a base model that requires further development.
- Letting the assistant write to accounting systems without review. Reports and drafts, yes; postings, pricing, and shipments only with employee confirmation.
- Failing to update the knowledge base. Outdated procedures produce confident but incorrect answers with links to old documents.
- Forgetting about personal data. Decide in advance which categories of data the assistant is allowed to process.
- Not planning an exit. Scenarios and a knowledge base built inside one ecosystem are harder to migrate.
Frequently asked questions
How much does Alice AI for business cost?
For Yandex 360 customers and new organizations, there is a three-month trial period during which the assistant is free. Current pricing after the trial is published on the product page in Yandex 360—check it before making a decision.
Can Alice AI for business be connected to 1C?
Yes, 1C is listed among the systems that can be connected in the “Plugins” section, along with Bitrix24, amoCRM, and SPARK. Which 1C objects the assistant can see and whether it can change them should be checked in a pilot.
Does the model train on company data?
According to Yandex’s statement on the product page, the organization’s data is not used to train the models and is stored in Yandex’s Russian data centers.
Can Alice AI be deployed in our own environment?
Yandex announced a customizable version with deployment in Yandex Cloud, in a hybrid setup, or on-premise. As of October 3, 2026, the launch conditions have not been published in official sources.
What is Alice AI Foundation, and how is it different from the assistant?
It is an open base language model with 80 billion parameters (3 billion active) under the Apache 2.0 license. The assistant is a ready-to-use service; Foundation is the raw material for building your own AI product, which needs to be further trained.
Can Alice AI Foundation be used in a commercial product?
Yes, the Apache 2.0 license permits commercial use. Responsibility for customization, answer quality, security, and legal compliance rests with the company building the service.
Is Alice AI for business suitable for a small company?
Yes, if the company already uses Yandex 360 or is ready to use it. It’s best to start with one process and a measurable goal, not by connecting all employees.
Primary sources and verification method
All facts were verified on October 3, 2026. We did not test the assistant ourselves: the review is based on official Yandex materials and business media, and the selection framework and pilot plan are editorial analysis by AIrassvet.
- Yandex press release dated September 8, 2026 — launch, modes, integrations, 185,000 organizations, trial period, announcement of a customizable version.
- Alice AI for Business page in Yandex 360 — department-by-department scenarios, knowledge base, data protection measures.
- AliceAI-Foundation-80B-A3B-Base card on Hugging Face — architecture, license, benchmarks.
- Technical breakdown of the model on Habr — training and architecture described by the Yandex team.
- T-Business Secrets: AI News for September 2026 — launch context, NeuroLawyer, timeline for 243-FZ regulations.
How AIrassvet helps implement an AI assistant in company processes
A ready-made assistant handles common employee tasks, but it doesn’t decide which process to automate first or how to connect AI to an accounting system without risking data. AIrassvet works with enterprise AI spaces, intelligent AI agents integrated with business systems, RAG systems based on corporate knowledge, and LLM deployment, including on-premise. When it comes to choosing between Alice AI for business and a custom system, we can:
- audit business processes and determine which tasks a ready-made assistant can handle and which require a custom integration;
- prepare a knowledge base — procedures and instructions that the assistant or your own RAG system will use to answer with source links;
- build an AI agent integrated with 1C or a CRM, with limited permissions and employee confirmation for critical actions;
- deploy a language model in your own environment, if the data cannot be sent to an external cloud.
A realistic first step is to choose one process and document its current state, data sources, constraints, and acceptance criteria. After that, it becomes clear whether a ready-made assistant is enough or whether you need a custom system.
Bottom line: Alice AI is a fast start, but not the whole architecture
Alice AI for business is a significant step for the Russian market: an agentic assistant built into familiar services, with plugins for 1C, Bitrix24, and amoCRM, data stored in Russian data centers, and a free three-month pilot. For companies on Yandex 360, it’s the best way to quickly test where AI can save employee time.
But a ready-made cloud assistant is not the company’s entire AI architecture. Critical processes that write to accounting systems, data that cannot leave the environment, and proprietary customer services require a separate solution: a customizable version after launch, or your own system, including one built on the open Alice AI Foundation. Start with one process, measure results during the trial period, and then decide what to scale based on the data.