AI glossary for business

LLM (Large Language Model)

The foundation for text generation, classification and dialogue. In business, LLM doesn't replace CRM — it adds a layer for unstructured data: emails, documents, chats.

Model choice depends on language, compliance and budget: Claude, OpenAI, YandexGPT, GigaChat, self-hosted.

RAG (Retrieval-Augmented Generation)

LLM answers based on your documents: retrieve relevant chunks → inject into prompt → answer with source citation.

RAG reduces hallucinations and is essential for support, knowledge bases and internal assistants.

AI agent

A system that not only replies but acts: creates CRM deals, generates PDFs, triggers workflows, escalates to humans.

Agent = LLM + tools (API, MCP) + security policies + monitoring.

MCP (Model Context Protocol)

Standard for connecting LLMs to external systems: CRM, databases, file storage. MCP server describes tools the model can invoke.

Simplifies agent integration with corporate IT without chaotic custom plugins.

LangGraph

Framework for multi-step agents with state, branching, human-in-the-loop and observability — when scenarios exceed simple prompt chains.

On-prem / Self-hosted AI

Deploying models and AI services inside company perimeter. Data never goes to public APIs. Required for banks, government and strict compliance.

Guardrails

Security policies around LLM: PII filtering, blocking dangerous actions, token limits, escalation on low confidence, request audit.

Claude AI — what is it and why does business need it?

Claude AI is Anthropic's LLM for text, documents, code, and multi-step tasks. In business it's used as an automation engine — not just a chat: proposal generation, support replies, document extraction, CRM updates.

Claude API — connecting to CRM, ERP, Bitrix24

Claude API (Anthropic API) is called from your systems: amoCRM, Bitrix24, 1C, n8n, Telegram, email. Keys, proxy if needed, rate limits, logging, and retries are configured. API keys and infrastructure stay with the client.

Claude Code

Claude Code is Anthropic's agentic coding tool for terminal and IDE workflows. Companies use it to accelerate integrations, scripts, MCP servers, and internal tools.

Claude MCP

Claude MCP (Model Context Protocol) connects Claude to CRM, databases, files, and internal APIs. MCP servers let agents pull data from amoCRM, 1C, or Confluence — without copy-paste and manual exports.

Claude in Russia

Most searches are about site and API access. For companies a working perimeter is designed: proxy, corporate keys, on-prem, or hybrid with YandexGPT/GigaChat where compliance requires it. Data stays in the client's environment.

Claude agent for sales and support

A Claude agent can handle inbound leads, draft commercial proposals, update CRM, answer from a knowledge base, and escalate to humans. Typical cases: sales ops, support, document workflows.

Claude Opus, Sonnet, and Pro

Opus for complex documents and multi-step agents. Sonnet for CRM and support balance of speed and quality. Pro is Anthropic's team subscription. In production the model is picked by task, language, budget, and data requirements.

Related pages

• RAG & search →
• AI agents →