Streaming background agent output to SSE clients
AgentTools
Hey, I'm building a feature where I dispatch multiple agents in parallel from a tool call.
Each agent runs in the background with its own thread/session, and users can view each agent's progress in a separate route (/audit/:sessionId) via SSE.
The challenge: The dispatch tool returns immediately (fire-and-forget), while the agents stream in the background. SSE clients connect after the agents have started streaming. I need to pipe each agent's stream to its corresponding SSE endpoint.
Current approach: I'm using a Node.js EventEmitter to bridge the gap:
This works, but feels like I'm reinventing something that might already exist.
Question: Is there a more "Mastra-native" way to handle this? Something like:
- Passing a custom writer/sink to agent.stream()
- A built-in way to tee/pipe agent streams to multiple destinations
- A streaming primitive designed for background agents with late-subscribing consumers
Or is the EventEmitter/pub-sub pattern the expected approach for this use case?
Using latest version of mastra and ai
Each agent runs in the background with its own thread/session, and users can view each agent's progress in a separate route (/audit/:sessionId) via SSE.
The challenge: The dispatch tool returns immediately (fire-and-forget), while the agents stream in the background. SSE clients connect after the agents have started streaming. I need to pipe each agent's stream to its corresponding SSE endpoint.
Current approach: I'm using a Node.js EventEmitter to bridge the gap:
This works, but feels like I'm reinventing something that might already exist.
Question: Is there a more "Mastra-native" way to handle this? Something like:
- Passing a custom writer/sink to agent.stream()
- A built-in way to tee/pipe agent streams to multiple destinations
- A streaming primitive designed for background agents with late-subscribing consumers
Or is the EventEmitter/pub-sub pattern the expected approach for this use case?
Using latest version of mastra and ai
