What is Gleyk
A platform for corporate knowledge and AI agents: knowledge bases, assistants with sources, agents on the canvas, automations, 1C and deployment inside your perimeter.
Gleyk is a platform for managing corporate knowledge, AI assistants and AI agents. It collects knowledge from documents and work systems, builds a single search index, answers questions with links to the original sources and performs permitted actions in your systems. Gleyk is not a language model: the platform connects local or external models and manages data, search, assistants, agents, connectors, access rights and logs.
What the platform consists of
- Knowledge bases. Documents, file storage, email, chats and corporate systems are loaded through connectors or manually, split into chunks and indexed. Each base has its own access rights and its own embedding model.
- Assistants and chat with sources. A user asks a question, and the assistant searches the connected bases and answers with references to document fragments that can be opened and checked. You can attach files to a question and ask for a Word document.
- Agents on the canvas. A visual builder where a scenario is assembled from components: search, model, conditions, code, tool calls. An agent does not only answer, it also performs a formalized sequence of actions.
- Automations. An agent starts on a schedule, a webhook or a connector event, and the result is delivered to a chat, Telegram or an external system.
- Telegram bots. The assistant works in Telegram, and actions that require approval arrive with "Confirm" and "Reject" buttons.
- Research mode. For questions that require going through hundreds of records from several sources: the assistant builds a plan, runs tasks in the background, computes aggregates with code and returns with a report and a source legend.
- Connectors and 1C. 1C bases are connected through OData: an agent reads documents, reference lists and registers, and prepares drafts and summaries. Connectors are available for Confluence, Jira, SharePoint, OneDrive, Google Drive, Outlook, IMAP, Zulip, Teams and other systems.
- API, SDK and MCP. Everything available in the interface is available programmatically: an HTTP API with a key, a Python SDK, an embeddable chat and a registry of MCP servers for agent tools.
Assistant and agent
These are two different mechanisms. An assistant is the entry point for a user: a dialogue, a question, a knowledge search, an answer with sources. It decides on its own whether search is enough, whether an external tool is needed or whether to run an agent scenario.
An agent is an executing component for a specific task or business process: analyzing a document, handling a request, classifying an inquiry, preparing a report, acting in an external system. An agent is assembled on the canvas from blocks and can be called by an assistant, a user, a schedule or an event.
Where the model runs
The model is a replaceable component. Placement options:
| Option | Where requests are processed | When to choose it |
|---|---|---|
| Private model in your perimeter | On your own GPU servers through vLLM or another OpenAI-compatible server | Sensitive data, an isolated perimeter, a high request volume |
| External provider | At the chosen provider, under configured data transfer rules | Individual tasks where sending data outside is acceptable |
| Managed option | On Gleyk resources in a dedicated environment | You have no GPU infrastructure of your own |
The platform itself is installed inside your perimeter: on a single server with Docker Compose or in a Kubernetes cluster, including on Astra Linux and RED OS. A protection layer works between the platform and the model: request checks, masking of sensitive data and a provider allowlist.
Note
An assistant's answer remains the result of automated processing. The platform provides sources precisely so that you can verify the conclusion before making a decision.
Where to go next
Choose a path by your role:
- User: how to ask questions, read answers with sources, work with documents and run research. Start with the user path and the article Chat with sources.
- Administrator: installation in your perimeter, organizations and roles, SSO, connecting 1C and other sources. Start with the administrator path and the wizard Deploy Gleyk on-prem.
- Developer: API key and SDK, embedding the chat on a website, MCP servers, agents on the canvas. Start with the developer path and the article API key and SDK.
The overall structure of the platform is described in the article Architecture. The official guides on installation, operation and functional characteristics are collected in the Documents section.
Ask Sema