Что такое Sloy простыми словами?

Слой корпоративной памяти: разрозненный рабочий след - переписки, встречи, документы, задачи, решения - связывается с проектами и клиентами так, чтобы ИИ-агент отвечал по нему без ручного пересказа.

Зачем это, если есть общий диск и мессенджер?

Диск хранит файлы, мессенджер хранит сообщения. Ни то, ни другое не отвечает на вопрос «что мы решили по этому клиенту и почему». Sloy связывает след с сущностями бизнеса, поэтому решение находится вместе с его контекстом.

Кто видит эти данные?

Доступ настраивается по источникам и проектам; агент читает то же, что разрешено человеку в его роли. Подключение начинается с одного реального проекта или папки, а не со всего контура компании.

Как это связано с бухгалтерией?

Агент, который проводит документы, должен знать контекст клиента: договорённости, особенности, прошлые исключения. Sloy отдаёт этот контекст, а ИИ-бухгалтер делает саму учётную работу.

Sloy — Corporate Memory for AI Agents

Sloy makes the company accessible to AI agents: the work trail from chats, meetings, documents, tasks, code and finance turns into verifiable memory across projects, people and decisions.

The company becomes accessible to agents

Sloy connects sources, memory, policies, finance and action

01

Sources

chats, email, Drive, Plaud, Jira, Git, 1C, banks

02

Memory

projects, clients, employees, decisions, monthly summaries

03

Policies

shared vocabulary: Epic, L30, TTU, MARGIN, IPR

04

Finance

P&L, cash flow, cash flow budget, plan-vs-actual, margin, payments

05

AI agents

understand the context, ask questions, prepare actions

06

Action

reply to the client, control, posting, escalation, decision

Corporate Memory

Conversations, documents and tasks become a circuit of understanding

People don't change their familiar systems

Google Docs, CRM, task trackers, email and conversations keep living where they're convenient for people. Sloy captures the work trail and links it to projects, clients, employees and decisions.

Memory becomes machine-readable

READMEs, short markdown files, change history, monthly summaries and links to sources are stored in the corporate GIT. Such context is easy for an agent to read and for a human to verify.

The original remains the source of truth

Google Docs, Sheets, PDFs, presentations and transcripts are stored as sources. The agent reads a lightweight version and returns to the original to verify a disputed fragment.

The context belongs to the company

Knowledge no longer stays in employees' private chats and is not lost when people change. The company sees what happened, who made the decision and on which facts it was based.

Wiki instead of RAG

Knowledge is compiled into an llm-wiki instead of being retrieved anew for every question

No RAG

Sloy doesn't search for random chunks at answer time. The context is turned in advance into a connected set of short markdown representations.

Source of truth

Original documents, emails, meetings and tasks remain immutable sources of truth.

Context access

The agent first reads the project map, monthly summaries and relevant pages, so it sees the accumulated synthesis.

Anti-hallucination

The answer relies on verifiable pages and links to sources, while contradictions are recorded in the project memory.

A question instantly assembles the context

The company gets answers that previously required manually collecting chats and files

What's the status of project A?

The agent gathers the latest decisions, risks, open questions, owners and the next step.

Where did the client approve this?

The answer includes the email, message, meeting, document or task where the confirmation appeared.

Where did the margin drop?

Sloy compares plan vs. actual, revenue, labor costs, expenses and agreements.

What happened over the month?

The system gathers completed work, disputed points, assignments and the money dynamics.

We support specific sources, not an abstract “knowledge base”

Sloy connects work traces where they already live. All data for each project is stored in the corporate GIT, and external repositories can be connected to the project. Access is granted on behalf of a user or organization—there is no need to add a bot to every chat.

Chats: conversations are linked to projects and people

Groups, channels, direct work chats, and client conversations. Context is linked to projects and people.

Email: messages become part of the decision history

Incoming emails, attachments, approval chains, client emails, and internal decisions.

Files: files and meetings become lightweight memory

Folders, meeting transcripts, documents, spreadsheets, and audio notes. Every document gets a lightweight AI representation.

Work: tasks show implementation progress

Deals, tasks, statuses, comments, SLAs, and implementation history. New connectors can be added quickly.

Finance: financial data answers questions about performance

Management accounting, 1C, spreadsheets, financial reports, and external financial sources for questions about project performance.

Code: code and documentation in the same environment

Project memory in the corporate GIT: sources, summaries, decisions, and links. External repositories connect as separate sources.

People: employee roles and workload are visible

Each employee has their own workspace: role, meetings, workload, feedback, development, and overload risks.

Extensions: add a new source without rebuilding the product

If a system provides data through an API, export, or webhook, we connect it as a new context source.

The original remains the source of truth; the agent reads a lightweight version

Each document is stored in two forms: the original as the source of truth and a lightweight representation for agents to read. The lightweight version contains structure, decisions, tables, entities, dates, people, finances, risks, and links to the original.

Google Docs provide decisions, not a wall of text

Headings, decisions, agreements, open questions, and links to exact locations instead of rereading the entire document.

Tables as meaning, not a grid

Sheets, ranges, key metrics, formulas, and conclusions—with context beyond blank cells and formatting.

PDFs and presentations become part of the work context

Documents are broken down into semantic blocks, tables, captioned images, and citable facts.

Meetings become memory immediately

The transcript is condensed into decisions, action items, risks, participants, and links to a project or employee.

Executives see risks and potential across the entire portfolio

Portfolio radar: where tension is rising, where upsells are possible, and which problems recur across clients. For example: “Which project has the most strained client relationship?” Sloy compares message tone, response delays, complaints, cancelled meetings, postponed payments, and unresolved promises.

Which projects have upsell potential?

The agent finds clients where trust already exists, requests recur beyond the contract, processes are only partly automated, volume is growing, and there is a clear financial opportunity to expand.

Which problems do clients mention most often?

Sloy groups complaints and signals by topic: response speed, status quality, documents, waiting for decisions, unclear invoices, and stalled approvals.

What significant events occurred during the vacation?

For a project manager, developer, or accountant, the agent gathers only what matters: decisions, risks, schedule changes, agreements, escalations, payments, blockers, and who needs a response.

One context layer for executives, PMs, and the team

For the executive: the full portfolio

Risks, upsell potential, recurring problems, and projects requiring management attention.

For the project manager: history without manual collection

Decisions, promises, client sentiment, deadlines, and financial context without manually collecting status updates.

For the project team: quick context access

Knowledge transfer, return from leave, and understanding what changed—without recapping everything in chats.

Assess where AI can deliver impact in your process

clients@kt.team Telegram @kt_team_it

Test Sloy on a real project—in one day

In one day, we connect one real environment: a project, client, folder, Telegram group, email, Drive, Plaud, or another source. The pilot shows how a fragmented work trail becomes corporate memory and how an agent gets context without manual retelling.

In the cloud Sloy validates value quickly: connect the first sources and see how work context becomes memory. Within your environment Sloy runs on the customer's servers—for companies that need to keep conversations, files, meetings, Git context, finances, and MCP access within their own infrastructure.

The article explains how the llm-wiki approach differs from RAG, knowledge graphs, and fine-tuning. "How to Give an LLM Your Knowledge". If you need a traditional document-search knowledge base, start with the page RAG.

New sources flow into a shared intake layer and route themselves

01

An email or file enters the shared context

An email, Plaud recording, Drive file, or chat enters the shared intake context.

02

The LLM suggests a project based on participants and text

The LLM reviews the text, participants, links, folder, and history of similar sources, then suggests a project.

03

A person confirms when confidence is low

When in doubt, Sloy asks the folder, client, business area, or employee owner.

04

Similar sources route themselves

Similar future emails, files, and meetings are routed automatically without manual sorting.

05

A heavy document gets a concise representation

Structure, facts, tables, decisions, and links to the original source.

06

The monthly summary preserves money, decisions, and risks

For every transport operation and project: finances, decisions, risks, and open questions.

07

The agent reads the relevant memory, not the archive

The agent reads the relevant monthly and project memory. Raw sources are available when clarification is needed.

08

A project is created from a request

"What is the status of Project N?" opens the project, retrieves its sources, and builds the first layer of memory.

09

Each employee is maintained as a workspace with access rules

For HR, mentoring, and audits, each employee is managed as a separate project type.

Sloy is needed wherever decisions are made in conversations, meetings, and tasks

Accounting is just one use case. The same layer is needed by service companies, implementers, lawyers, consultants, support, sales, HR, and internal teams. Each project's context lives in the corporate GIT and is available through MCP—read by Codex, Claude, Cursor, Openclaw, Hermes, and other agents without vendor lock-in.

The client gets a status update without asking the manager

A client asks for a status update in Telegram. The agent sees Jira, emails, meetings, and agreements, then prepares a response without asking the manager.

The project sees where a decision is stuck

Meetings, tasks, correspondence, and lightweight document versions are assembled into a project. You can see where decisions are stuck and who needs to respond.

Financial data is read with its context

"What is the financial trend?" reads management reports, 1C, spreadsheets, and monthly summaries—not the entire message archive.

Document, email, and accounting entry in one trail

The document, client email, manager's comment, and accounting entry are linked into one context for review and month-end closing.

Employee overload becomes visible before burnout

The employee as a project: workload, meetings, feedback, conflicts, growth, and early signs of overload.

Request history speeds up responses

Request history, SLAs, emails, chats, and tasks give the agent context for responses and escalations.

Promises to clients are not lost after meetings

Bitrix24, Gmail, meetings, and chats show what was promised to the client, who makes the decision, and what the next step is.

One prompt instead of manual configuration

Sloy provides text for the agent environment: the user opens an empty folder, pastes a prompt, and the agent registers MCP access to the required project.

Without Sloy, the agent sees a message; with Sloy, it sees the working situation

	Without Sloy	With Sloy
What the agent sees	Only the current chat: every new agent asks again for an explanation of the client, project, and decision history.	Project memory in the corporate Git: sources, monthly summaries, decisions, finances, meetings, and lightweight document versions.
Sources	Scattered across chats, folders, CRM, email, tasks, and people’s heads	Connected and linked to projects
Project	Reconstructed manually	Routes itself
AI agent	Reads the current chat	Works with project memory