Session 8· 06· 15 min

Streaming Agent Output

What you'll learn
  • Stream agent progress with stream_mode="updates"
  • Stream individual tokens with stream_mode="messages"
  • See tool calls appear in real time

Agents may take several seconds for multi-step reasoning. Streaming shows progress in real time — users see the model thinking, tools executing, and the answer arriving token by token.

stream_mode="updates"
step progress
  • One event per agent step
  • Shows tool calls and results
  • Good for progress indicators
stream_mode="messages"
token streaming
  • One event per token
  • Typewriter effect
  • Good for live chat UIs
$ python 06_streaming.py
Knowledge Check
Which stream mode would you use for a chat UI that shows tokens appearing one by one?
Recap — what you just learned
  • agent.stream() replaces agent.invoke() for real-time output
  • "updates" = one event per step; "messages" = one event per token
  • You can combine both: stream_mode=["updates", "messages"]
Next up: 07 — Structured Output