Agently Docs

EN

中文

EN

中文

Appearance

Resize & Strategy Extension ​

Applies to: 4.0.8+

resize governs:

  • context_window size
  • optional memo updates

1) Default flow: max_length + simple_cut ​

Built-in analyzer:

  • if session.max_length is int and window exceeds it -> strategy simple_cut

Built-in simple_cut executor:

  • keeps newest messages until total length <= max_length

Config:

python
agent.set_settings("session.max_length", 12000)

2) Automatic vs manual trigger ​

  • automatic: with auto_resize=True
  • manual: call session.resize()
python
agent.activate_session(session_id="resize_demo")
session = agent.activated_session

session.resize()
# await session.async_resize()

3) Custom strategy: keep last N turns ​

python
def analysis_handler(full_context, context_window, memo, session_settings):
    if len(context_window) > 10:
        return "keep_last_ten"
    return None


def keep_last_ten(full_context, context_window, memo, session_settings):
    return None, list(context_window[-10:]), memo


session.register_analysis_handler(analysis_handler)
session.register_execution_handlers("keep_last_ten", keep_last_ten)

4) When to disable auto_resize ​

For batch writes where you want one explicit compaction pass:

python
from agently.core import Session

session = Session(auto_resize=False, settings=agent.settings)
# write many messages
session.resize()

5) Common pitfalls ​

  • setting session.max_length without activating a session
  • returning an unregistered strategy name
  • returning invalid executor tuple shape