Agently Docs

EN

中文

EN

中文

Appearance

Session Quickstart ​

Applies to: 4.0.8+

This page walks through the practical flow in v4.0.8:

  1. turn session on/off
  2. isolate sessions by id
  3. control recording fields
  4. export and restore
  5. customize window/memo strategy

1) Turn session on/off ​

python
from agently import Agently

agent = Agently.create_agent()

agent.activate_session(session_id="demo_on_off")
agent.input("Remember: buy eggs tomorrow.").streaming_print()
agent.input("What should I buy tomorrow?").streaming_print()

agent.deactivate_session()
agent.input("What should I buy tomorrow?").streaming_print()

Notes:

  • activate_session() auto-generates an id if omitted
  • same id reuses history, different ids isolate history

2) Session isolation by id ​

python
agent.activate_session(session_id="trip_a")
agent.input("Remember trip A destination: Tokyo.").streaming_print()

agent.activate_session(session_id="trip_b")
agent.input("Remember trip B destination: Paris.").streaming_print()

agent.input("What is my destination?").streaming_print()  # trip_b

agent.activate_session(session_id="trip_a")
agent.input("What is my destination?").streaming_print()  # trip_a

For web apps, use stable keys (user_id, tenant:user_id) as session_id.

3) Selective input/output recording ​

By default, full prompt/result is recorded. To record only key fields:

python
agent.activate_session(session_id="record_demo")
agent.set_settings("session.input_keys", ["info.task", "input.lang"])
agent.set_settings("session.reply_keys", ["summary", "keywords"])

result = (
    agent
    .info({"task": "Summarize Agently", "style": "technical"})
    .input({"lang": "en", "noise": "ignored"})
    .output({"summary": (str,), "keywords": [(str,)]})
    .get_data()
)

print(result)
print(agent.activated_session.full_context)

agent.set_settings("session.input_keys", None)
agent.set_settings("session.reply_keys", None)

4) Export and restore ​

python
from agently.core import Session

agent.activate_session(session_id="export_demo")
agent.input("Remember recovery code: X-2025-ABCD").streaming_print()

exported = agent.activated_session.get_json_session()

restored = Session(settings=agent.settings)
restored.load_json_session(exported)
restored.id = "export_demo_restored"
agent.sessions[restored.id] = restored
agent.activate_session(session_id=restored.id)

agent.input("What recovery code did I ask you to remember?").streaming_print()

5) Custom window strategy and memo ​

python
def analysis_handler(full_context, context_window, memo, session_settings):
    if len(context_window) > 4:
        return "keep_last_four"
    return None

async def keep_last_four(full_context, context_window, memo, session_settings):
    kept = list(context_window[-4:])
    new_memo = {"kept_count": len(kept)}
    return None, kept, new_memo

agent.activate_session(session_id="memo_demo")
session = agent.activated_session
session.register_analysis_handler(analysis_handler)
session.register_execution_handlers("keep_last_four", keep_last_four)

Continue with: