AGTerm

Multi-model AI terminal agent with context management, memory compression, and autonomous execution.

Overview

AGTerm is a powerful AI terminal agent in the Calliope platform suite. It provides:

Secure terminal server - Web-based terminal access with authentication
AI agent integration - Multiple AI backends including Claude Code, Gemini CLI, and custom agents
Context engine - Intelligent context assembly with token budgeting
Memory management - Automatic conversation compression
Multi-model gateway - Unified access to 12+ AI providers

AGTerm is available as a tool in The Hub alongside:

AI IDE - Full development environment with AI
AI Lab - JupyterLab with AI assistance
Chat - Conversational AI interface
Deep Data Agent - Autonomous research agent
Agentic Browser - Web automation agent

Architecture

┌─────────────────────────────────────────────────────────────┐
│                      AGTerm Server                          │
├─────────────────────────────────────────────────────────────┤
│  ┌─────────────┐  ┌──────────────┐  ┌───────────────────┐   │
│  │   Session   │  │   Context    │  │      Memory       │   │
│  │   Manager   │  │    Engine    │  │     Manager       │   │
│  └─────────────┘  └──────────────┘  └───────────────────┘   │
│                                                             │
│  ┌─────────────────────────────────────────────────────┐   │
│  │                    Agent Core                        │   │
│  │  ┌─────────┐ ┌─────────┐ ┌─────────┐ ┌─────────┐    │   │
│  │  │ Claude  │ │ Gemini  │ │ Codex   │ │ AGTerm  │    │   │
│  │  │  Code   │ │   CLI   │ │   CLI   │ │  Agent  │    │   │
│  │  └─────────┘ └─────────┘ └─────────┘ └─────────┘    │   │
│  └─────────────────────────────────────────────────────┘   │
│                                                             │
│  ┌─────────────────────────────────────────────────────┐   │
│  │                Multi-Model Gateway                   │   │
│  │  Anthropic | OpenAI | Google | Mistral | Ollama ... │   │
│  └─────────────────────────────────────────────────────┘   │
└─────────────────────────────────────────────────────────────┘
          │                            │
          ▼                            ▼
    ┌───────────┐              ┌───────────────┐
    │  Web UI   │              │  WebSocket    │
    │  Client   │              │  Connections  │
    └───────────┘              └───────────────┘

Key Features

Session Management

Create and manage terminal sessions:

Multiple sessions - Run multiple AI agents simultaneously
Session persistence - Sessions survive disconnects
Session switching - Seamlessly switch between agents
Workspace isolation - Each session has its own working directory

AI Backends

Switch between multiple AI backends:

Backend	Description
AGTerm Agent	Native multi-model agent with tool use
Claude Code	Anthropic’s coding assistant
Gemini CLI	Google’s AI CLI
Codex CLI	OpenAI’s code model
Shell	Plain terminal (no AI)

Context Engine

Intelligent context assembly for AI conversations:

Token budgeting - Stay within model limits
Priority pruning - Remove low-priority content first
Explicit files - Add specific files to context
Folder scanning - Include project structure
Search integration - Add search results to context

Memory Management

Automatic conversation memory with compression:

Auto-compression - Compress old messages when approaching limits
LLM summarization - Use AI to create intelligent summaries
Entry tracking - Message, tool call, tool result, context entries
Import/export - Save and restore session state

Multi-Model Gateway

Unified interface to multiple AI providers:

Provider	Models
Anthropic	Claude Sonnet 4, Opus, Haiku
OpenAI	GPT-4o, GPT-4 Turbo
Google	Gemini 2.0, Gemini 1.5
Mistral	Large, Codestral
Groq	Llama 3.3
Together	Open source models
OpenRouter	Any model
Ollama	Local models