An AI agent that lives where you work β€” terminal, email, chat, or SSH

Install Configuration GitHub

Multi-Transport

Terminal, SSH, email, MQTT β€” one agent through any door.

Rich Toolset

Shell commands, browser automation, MCP tools, webhooks β€” extensible.

Multi-Agent

Dispatch sub-agents in parallel for code review, security audit, and more.

Skill System

Markdown files that teach β€” load on demand, keep prompt lean.

Scheduled Tasks

CRONTAB.md β€” define recurring tasks in cron syntax.

Session Persistence

Checkpoints, summaries, diary β€” agent state never lost.

Dolphin is a cross-platform AI agent that runs shell commands, controls a browser, delegates work to sub-agents, and follows schedules you define. It connects through whatever channel suits the task β€” local terminal, SSH, MQTT, or email.

Why Dolphin?

Most AI coding tools lock you into a specific editor or a web UI. That’s fine for writing code, but real work sprawls. You might want to ask the agent something over email while you’re on your phone. Or have it run a scheduled task every evening without anyone touching a keyboard. Or SSH into a server and ask the agent sitting there to diagnose an issue.

Dolphin doesn’t care which door you knock on β€” it answers them all. The same agent, the same tools, the same session state, regardless of transport.

Quick Start

# Set your API key and provider
export DZ_LLM_TYPE="openai"
export DZ_LLM_API_KEY="sk-..."
export DZ_LLM_BASE_URL="https://api.openai.com/v1"
export DZ_LLM_MODEL="gpt-4o"

# Run dolphin
./dolphin

On the first run, Dolphin will walk you through a setup wizard β€” choose your role, optionally generate a config file and a system prompt file. See the Quick Start guide for other providers.

Key Features

Distribution