Container images

You already have an agent, written your way? Deploy its container image instead of a harness. Waken turns the image into a persistent machine: your program keeps its disk and its memory, sleeps when idle, wakes on the next message, and can be rewound like any other agent.

The contract

Your program listens on the port given in $PORT (18789 by default) and answers two routes:

• GET /health: answer 200 when you are ready.
• POST /chat with {"message": "...", "source": "api"}: answer with plain text, or JSON with one of response, reply, message, text or output.

That is the same contract as Maritime: an agent written for it runs on Waken without changes.

import os, json
from http.server import BaseHTTPRequestHandler, HTTPServer
 
class Agent(BaseHTTPRequestHandler):
    def do_GET(self):
        self.send_response(200); self.end_headers(); self.wfile.write(b"ok")
    def do_POST(self):
        body = json.loads(self.rfile.read(int(self.headers["Content-Length"])))
        reply = json.dumps({"response": "You said: " + body["message"]}).encode()
        self.send_response(200); self.end_headers(); self.wfile.write(reply)
 
HTTPServer(("0.0.0.0", int(os.environ.get("PORT", 18789))), Agent).serve_forever()

Deploy

curl -X POST https://api.waken.sh/agents \
  -H "Authorization: Bearer $WAKEN_KEY" \
  -d '{
    "name": "my-agent",
    "image": "ghcr.io/you/my-agent:1.4",
    "env": { "OPENAI_API_KEY": "sk-..." }
  }'

Optional: command replaces the image's entrypoint and command, and port changes the port. The first deploy of an image takes a few seconds while it is downloaded. After that, a new agent from the same image starts in under two seconds.

Then talk to it like any agent, with POST /agents/{id}/message. Files, commands, logs, variables, schedules and checkpoints all work the same. Keep what must survive in /data or anywhere on disk: the whole disk persists.