Skip to content

Steering & Follow-up

Steering and follow-up queues let you inject messages into a running agent's conversation. Steering messages are delivered between turns; follow-up messages are delivered after the current run completes.

Overview

  • Steering -- Use agent.steer(message) to inject a message between turns. The agent sees it before its next LLM call.
  • Follow-up -- Use agent.follow_up(message) to queue a message for after the current run. The agent processes it as a new prompt.

Queue modes

Both queues support two drain modes:

Mode Behavior
"one-at-a-time" Process one queued message per turn (default)
"all" Process all queued messages at once 
agent.steering_mode = "all"
agent.follow_up_mode = "one-at-a-time"

steer()

agent.steer(UserMessage(
    content="Focus on cost analysis.",
    timestamp=int(time.time() * 1000),
))

follow_up()

agent.follow_up(UserMessage(
    content="Now summarize your findings.",
    timestamp=int(time.time() * 1000),
))

Queue management

Content coming soon.

Next steps