The OI Architecture

Building a sovereign interface for agent-orchestrated operations.

System Overview

TerminaI is designed as a modular **Operator Interface (OI)**. Unlike coding copilots which are integrated into IDEs, TerminaI is the first AI that sits directly at the system level—controlling the shell, the GUI, and remote agents.

Monorepo Structure

  • @google/gemini-cli (Original Fork)
  • @terminal/core (Agentic Harness)
  • @terminal/a2a (Communication Protocol)
  • @terminal/voice (Local STT/TTS)

The Stack

  • • Node.js & TypeScript for Infrastructure
  • • Tree-sitter for Syntax Understanding
  • • Fastify for High-Performance RPC
  • • Framer Motion for UI Experience

Architecture Diagram

TerminaI architecture stack

The stack keeps execution local, layers approvals for every action, and exposes secure protocols for MCP and A2A orchestration.

The Execution Stack

The Host (PTY)

TerminaI doesn't wrap subprocesses; it spawns a true pseudo-terminal. This allows for native colors, TTY signals, and full terminal interactive capability.

node-pty, xterm.js

The Thinking Core

An agentic harness built on Gemini 2.5 Pro. It handles multi-step planning, state management, and error correction during execution.

Gemini 2.5 Pro, 2M Context

The Valve (Safety)

Every command passes through our three-tier Approval Ladder. Policies are locally enforced and cannot be overridden by the model.

Local Policy Engine, PIN Auth

MCP Host

A standardized interface for connecting models to local and remote data. Connect to GitHub, Slack, databases, and more via a unified API.

Model Context Protocol v1

A2A Protocol

The Agent-to-Agent protocol allows TerminaI to delegate tasks to other agents or be remotely controlled via our secure cloud relay.

JSON-RPC over HTTP/WSS

Built for Sovereignty

Every design choice in TerminaI—from local PTY execution to our offline voice stack—is focused on ensuring you remain the sovereign of your own system.

Explore the GitHub