Introduction

OpenAgent is an open-source, enterprise-grade AI agent platform built for teams that want to deploy powerful AI capabilities on their own infrastructure.

What can you build with OpenAgent?

• AI Knowledge Assistants — Upload your documentation, PDFs, and data. Let your agents answer questions with accurate, cited answers from your own knowledge base.
• Multi-Channel Bots — Connect a single agent to Telegram, Discord, Slack, and more simultaneously.
• Workflow Agents — Agents with MCP tool support that call APIs, execute code, search the web, and complete multi-step tasks autonomously.
• Enterprise Chatbots — Role-based access control, SSO integration, and audit logs for compliance-sensitive deployments.

Architecture Overview

OpenAgent is structured around four layers:

┌─────────────────────────────────────────┐
│              Channels Layer             │
│  Telegram · Discord · Slack · Web Chat  │
├─────────────────────────────────────────┤
│               Agent Layer               │
│   Reasoning · Tool Use (MCP) · Memory   │
├─────────────────────────────────────────┤
│              Model Layer                │
│  OpenAI · Claude · Gemini · Ollama ...  │
├─────────────────────────────────────────┤
│           Knowledge Base Layer          │
│   Documents · Embeddings · Search       │
└─────────────────────────────────────────┘

Channels Layer

Incoming messages arrive from any supported channel (Telegram, Discord, Slack, a web widget, etc.). The gateway routes them to the appropriate agent and returns the response back to the originating channel.

Agent Layer

The agent is the reasoning core. It receives the user message and conversation history, decides whether to retrieve from the knowledge base, call a tool via MCP, or respond directly, then generates a response using the configured model.

Model Layer

OpenAgent integrates with 30+ LLM providers through a unified API. You can use cloud APIs (OpenAI, Anthropic, Google) or run local models (Ollama, LM Studio) — the agent logic stays the same regardless of which model powers it.

Knowledge Base Layer

Documents are uploaded, parsed, split into chunks, and embedded using a vector embedding model. At query time, the agent performs semantic search to retrieve the most relevant chunks as context.

Key Concepts