Skip to main content

Async Sub-Agents

Sometimes your agent says: "I'm working on that in the background — I'll get back to you." That's an async sub-agent at work.

Your primary agent stays with you, handling the conversation in real time. When it hits something slow — an outbound phone call, a prior authorization, a multi-step lookup — it spins up a second agent to do the work in the background. You keep chatting. The primary agent reports back the moment the background work is done.

What This Enables​

Async sub-agents turn one agent into a small team. A few examples:

Outbound outreach in parallel​

Ask your primary agent to reach every patient who hasn't completed onboarding. It can spawn one background agent per patient, each making its own call, and roll the results up for you:

Agent 1 called Alice — she's getting back to us today. Agent 2 left a voicemail for Bob. Agent 3 learned that Carol has moved and no longer needs services.

You didn't have to wait through any of the calls.

Prior authorization in the background​

A caller asks about a prescription. Your primary agent stays with them, answering other questions, while a background agent begins the prior auth request. When authorization comes back, the primary agent weaves the update into the conversation naturally — no context switch, no "please hold."

Multi-step research​

Need five patient records looked up and cross-referenced? Your agent can fan out five parallel sub-agents, each reading one record, and synthesize the combined result when they're all done.

What You'll See​

Async delegation feels like the agent simply works. Concretely, you'll notice:

  • Progress messages in your chat: "Let me start that in the background..." or "Still waiting on the pharmacy — I'll let you know."
  • Unsolicited follow-ups when the background work finishes: "Quick update on the prior auth you asked about earlier — approved."
  • Live status on demand. Ask "What's happening with the Ozempic request?" at any time and the primary agent will check the background task and give you the latest.
  • Clean cancellation. If you change your mind, just say "cancel the background work on Ozempic" and the agent will stop the sub-agent.

How It Relates To Your Existing Agents​

You don't configure anything new. The primary agent decides when a task is too slow to block on and delegates automatically, drawing from the same set of agents you've already built in your workspace. Think of it as one agent borrowing another agent for a specific piece of work — no new accounts, no new seats, no new phone numbers.

Each sub-agent runs in its own chat, which you can see in the Chats view if you want to follow along. But the primary conversation never pauses waiting for them.

Safeguards​

Sub-agents run with real teeth and real guardrails.

  • Delegation depth is capped. A primary agent can delegate to a sub-agent, which can delegate to another, up to a maximum of three levels. This prevents runaway chains.
  • Ownership is enforced. Only the agent (and the person) that started a background task can check on it, send it new instructions, or cancel it. Other agents in the workspace can't even see it exists.
  • You can always cancel. A single instruction — "stop the background work on X" — ends the sub-agent gracefully.
  • HIPAA-compliant audit logging. Every time a background task's contents are read, a PHI_RECORD_VIEWED audit event is written. The audit trail records that PHI was accessed and by whom, never the PHI itself.
  • Workspace isolation. Sub-agents never cross workspace boundaries. A background task spawned in one workspace can only read, write, and act inside that workspace.

Learn More​