WAKE is a simple, open standard that tells any AI agent how to surface its work to you — and how to wait for your direction before moving on.
AI agents work in the background — researching, writing, making decisions. But there's no standard way for them to tell you what they've done, ask for your input, or wait before doing something irreversible. So they either go silent, or they interrupt you constantly.
When an agent implements WAKE, it knows exactly when and how to check in with you. When you use a WAKE-compatible inbox, you see everything in one place — and your decisions get back to the agent immediately.
Your agent finishes a task, hits a decision point, or encounters something unexpected. Instead of going silent or blasting you with a notification, it sends a structured delivery — a short headline and summary — to your WAKE inbox.
Deliveries wait for you. You can check in every hour, or every day. The agent keeps working on other things. When you're ready, you see exactly what it did and what it needs from you.
You approve, reject, or redirect with optional instructions. That response goes straight back to the agent, which picks up exactly where it left off.
Three things you can tell an agent
Proceed as planned. The agent continues with its original approach.
Stop. Don't proceed with this. The agent reads your feedback and adjusts.
Here's how I want you to change course. Your instructions go directly to the agent.
Any agent that can make an HTTP request can implement WAKE in minutes. Any inbox that can serve two endpoints is WAKE-compatible. No vendor lock-in, no new dependencies.
Claude, ChatGPT, Cursor agents, n8n workflows, custom scripts — if it can make an HTTP request, it can implement WAKE. You share a URL or a prompt and you're done.
Every delivery is signed. Your API key travels in the header, never in the payload. Tampered messages are rejected before they reach you. The agent prompt handles all of this automatically.
WAKE is an MIT-licensed open protocol. No company owns it. Any tool can implement it. If you switch agents, switch inboxes, or build your own — the protocol travels with you.
WAKE doesn't care which AI you use or how your agents are built. If it can follow an instruction, it can implement the protocol.
WAKE is the protocol. GardenRoom is one place to receive deliveries — a hosted inbox built around the spec. You don't have to use it. But if you want something that works today, it's there.
A hosted inbox for your WAKE deliveries. One URL, one prompt to your agents, and everything they do comes to you in one place.
WAKE is a public specification. The schema is stable, versioned, and designed to last. Anyone can build a WAKE-compatible inbox. Anyone can publish a WAKE-compatible agent. The community decides what comes next.
Paste this URL directly into your agent's context window or system prompt. Your agent will read the full WAKE spec and immediately know how to send you deliveries.
If your agent can't read URLs, paste this prompt into its system instructions instead. It teaches WAKE, sets up the delivery behaviour, and handles security automatically.
[YOUR_WAKE_ENDPOINT] and [YOUR_AGENT_ID], then set WAKE_API_KEY as an environment variable. Your agent handles everything else.