Agents That Work While You Sleep
Most AI assistants are synchronous. You send a message, you wait for a response. If the task takes five minutes of real work — research, multiple API calls, file processing — you're sitting there watching a spinner.
Edward's orchestrator changes this. It spawns lightweight worker agents that execute tasks independently, with full access to Edward's tools and memory.
Worker Agents
A worker is a mini-Edward. It runs within the same process, has its own conversation thread, and can use any tool the main Edward can — search the web, send messages, execute code, read documents, schedule events.
When you ask Edward to do something complex, it can break the work into tasks and delegate them to workers. Each worker reports back when done, and their conversations appear in the sidebar so you can review what happened.
Scheduled Tasks
The orchestrator pairs naturally with Edward's scheduling system. You can tell Edward to do something tomorrow morning, next Tuesday, or every Monday at 9am. The scheduler fires the event, the orchestrator picks it up, and a worker executes it.
This is where it gets interesting. Edward can schedule his own follow-ups. “Remind me to check the deployment tomorrow” becomes an actual scheduled event that, when it fires, gives Edward full tool access to check the deployment and message you with the results.
Crash Recovery
Workers can fail. The server can restart. The orchestrator handles this with atomic task pickup using SELECT ... FOR UPDATE in PostgreSQL. On startup, it scans for tasks that were in progress when the system went down and resumes them.
Why Not Just Use a Queue?
You could bolt on Celery or Redis queues. But Edward's workers aren't running arbitrary functions — they're having conversations with tools. The unit of work is a chat_with_memory() call, which means each worker gets memory retrieval, tool access, and conversation persistence for free. The orchestrator is thin because it delegates to infrastructure that already exists.
The Result
You can ask Edward to research a topic, compile the results, and email you a summary — then close your laptop. The workers handle it asynchronously, and you get a push notification when it's done. It's not just a chatbot anymore. It's an assistant that actually works in the background.