Skip to main content

Architecture Overview

Orient is a self-hosted AI agent platform. Everything runs on your infrastructure — no cloud dependency, no data leaving your control.

System Architecture​

┌──────────────────────────────────────────────────────┐
│                    Your Machine                       │
│                                                      │
│  ┌─────────────┐  ┌──────────────┐  ┌─────────────┐ │
│  │  WhatsApp    │  │    Slack     │  │    CLI      │ │
│  │  Connector   │  │  Connector   │  │  Interface  │ │
│  └──────┬───────┘  └──────┬───────┘  └──────┬──────┘ │
│         │                 │                 │        │
│         └────────┬────────┴────────┬────────┘        │
│                  │                 │                  │
│           ┌──────▼───────┐  ┌─────▼──────┐          │
│           │  Message      │  │  Agent     │          │
│           │  Router       │  │  Engine    │          │
│           └──────┬───────┘  └─────┬──────┘          │
│                  │                │                   │
│           ┌──────▼────────────────▼──────┐           │
│           │        Tool Executor         │           │
│           │  (MCP Servers, Integrations) │           │
│           └──────────────┬───────────────┘           │
│                          │                           │
│           ┌──────────────▼───────────────┐           │
│           │     SQLite Database          │           │
│           │  (messages, config, state)   │           │
│           └──────────────────────────────┘           │
└──────────────────────────────────────────────────────┘

Data Flow​

  1. Message arrives — from WhatsApp, Slack, CLI, or any connected platform
  2. Router dispatches — identifies the user, platform, and conversation context
  3. Agent processes — the configured agent picks up the message, loads relevant context (memory, integrations, previous conversations)
  4. Tools execute — the agent calls tools as needed (calendar, Jira, file operations, MCP servers)
  5. Response sent — the result flows back to the originating platform

Key Components​

Message Connectors​

Each platform has a dedicated connector that handles authentication, message format conversion, and delivery. Connectors are stateless — all state lives in the database.

  • WhatsApp — uses the WhatsApp Web multi-device protocol
  • Slack — uses the Slack Bolt framework with Socket Mode
  • CLI — direct terminal interface for development and power users
  • MCP — Model Context Protocol for IDE integration

Agent Engine​

The agent engine manages conversation flow, context windows, and tool calling. Each agent can be configured independently with:

  • Custom system prompts
  • Specific tool permissions
  • Platform-specific behaviors
  • Memory and context settings

Tool Executor​

Tools are organized as MCP (Model Context Protocol) servers. This means Orient's tools are compatible with any MCP client, and you can connect external MCP servers to extend functionality.

Built-in tool categories:

  • Scheduling — create, update, delete scheduled messages and reminders
  • Calendar — Google Calendar integration for events and availability
  • Jira — ticket management, sprint queries, updates
  • Mini-Apps — generate and host lightweight web applications
  • File operations — read, write, and manage files
  • Webhooks — receive and process external events

Database​

Orient uses SQLite for all persistent storage:

  • Conversation history
  • Agent configuration
  • Scheduled messages
  • Integration credentials (encrypted)
  • Feature flags
  • Webhook configurations

SQLite was chosen for simplicity — no external database server needed. The entire database is a single file on your machine.

Self-Hosted Deployment​

Orient can be installed with a single command:

curl -fsSL https://orient.bot/install.sh | bash
orient start

This starts:

  • Orient server — the main application
  • Dashboard — web UI at localhost:4098

For Docker-based deployments, see the docker/ directory in the repository.

No external services are required. All AI inference uses your configured LLM provider (OpenAI, Anthropic, etc.) with your own API keys.

No Cloud Dependency​

Orient has zero cloud dependencies:

  • No telemetry or analytics
  • No license server
  • No remote configuration
  • No automatic updates
  • All data stays on your machine

The only external calls are to your configured LLM provider, and those are made directly from your infrastructure.