Protege

Appearance

Protege 🥚Email-native AI agent framework

Build AI agents you communicate with over email — like a colleague, not a chatbot.

Get Started
Why Email?

Email as the Interface

Your agent sends and receives real email. No proprietary chat protocol. Works with Gmail, Outlook, Apple Mail, or any SMTP client.

Local-First Runtime

Agent logic, memory, and data stay on your machine. You choose the LLM provider. Nothing is hosted unless you opt into the relay.

Extension-First Design

Swap LLM providers, add tools, hook into events, and customize context assembly — all through a simple JSON manifest.

Scheduled Responsibilities

Define recurring tasks in markdown with a cron schedule. Your agent executes them asynchronously and emails you the results.

What is Protege? ​

Protege is a framework for building AI agents that communicate over email. It gives you the primitives — a gateway, an inference harness, a tool system, memory, scheduling, and an extension architecture — and you assemble them into whatever agent your use case requires.

It works out of the box: protege init, configure a provider, and you have a working agent in minutes. But the real value is what you build on top of it. Every layer is designed to be extended or replaced — write custom tools, swap LLM providers, hook into lifecycle events, inject your own context at inference time. Protege is the foundation; the agent you build is yours.

Protege is a framework, not a product

If you want something that just works out of the box without writing code or learning an extension system, check out OpenClaw or ChatGPT. Protege is built for developers and builders who want full control over their agent's behaviors, integrations, context loading, deployment, and more.

You (Gmail/Outlook/etc.)
   │
   ▼
┌───────────────────────────────┐
│  Protege Gateway (local)      │
│  ┌─────────┐  ┌──────────────┐│
│  │ Harness │──│ LLM Provider ││
│  │ (tools, │  │ (OpenAI,     ││
│  │ context)│  │ Anthropic,   ││
│  └─────────┘  │ Gemini,      ││
│               │ Grok)        ││
│               └──────────────┘│
└───────────────────────────────┘
   │
   ▼
You receive a reply email

Why Email? ​

Most agent providers lock you into a proprietary chat interface or require 3rd party owned communication channel (e.g. WhatsApp, Discord, Telegram, Slack) to communcate. Protege takes a different approach: email is already the world's most widely deployed asynchronous messaging protocol. By building on SMTP, Protege gets several things for free:

  • Universal compatibility — your agent works with every email client that exists
  • Asynchronous by default — no need to sit and watch your agent think. Send a message, close your laptop, get a reply later
  • Thread-aware history — email threading gives agents conversational memory without building a second protocol
  • Agent-to-agent communication — agents can email each other using the same open protocol

How It Works ​

  1. You email your agent at its address (e.g., charlie@mail.protege.bot)
  2. The gateway receives the message via SMTP (directly or through the relay bridge)
  3. The harness assembles context — your agent's persona, memory, conversation history, and the incoming message
  4. The LLM generates a response, optionally calling tools (web search, file I/O, shell, sending emails)
  5. The agent replies by email back to you, continuing the thread

What Can You Build? ​

  • A customer support agent that receives support emails, looks up account data, troubleshoots issues, and replies — without a human in the loop
  • A sales assistant that qualifies inbound leads, answers product questions, and forwards hot prospects to your team
  • A research assistant that searches the web and emails you summaries on a schedule
  • A billing & invoicing agent that processes invoice requests, generates line items, and emails confirmations to customers
  • A network of collaborating agents — e.g., a triage agent routes incoming mail to specialized agents for support, sales, and ops

Next Steps ​