Agently Docs

EN

中文

EN

中文

Appearance

Export, Import, and Recovery ​

Applies to: 4.0.8.1+

Session has built-in JSON/YAML serialization, suitable for:

  • persisting sessions
  • cross-process recovery
  • runtime snapshot auditing

1. What gets serialized ​

How to read this diagram ​

  • Session serialization stores conversational state, not the whole runtime environment.
  • If an object depends on the current process, network connection, or closure scope, do not expect Session snapshots to recreate it.

2. Export ​

python
session_json = session.get_json_session()
session_yaml = session.get_yaml_session()

# aliases also work
session_json_2 = session.to_json()
session_yaml_2 = session.to_yaml()

Exported content includes:

  • id
  • auto_resize
  • full_context
  • context_window
  • memo
  • session_settings

3. Import (string or file path) ​

python
from agently.core import Session

session = Session(settings=agent.settings)
session.load_json_session(session_json)
session.load_yaml_session("./session.snapshot.yaml")

# aliases
session.load_json(session_json)
session.load_yaml("./session.snapshot.yaml")

4. Extract from a larger object ​

When session data is nested inside a larger JSON/YAML object, use session_key_path:

python
session.load_json_session(
    "./snapshot.json",
    session_key_path="payload.session",
)

5. Restore into the Agent session pool ​

python
restored = Session(settings=agent.settings)
restored.load_json_session("./session.snapshot.json")

agent.sessions[restored.id] = restored
agent.activate_session(session_id=restored.id)

6. Compatibility and validation advice ​

  • record a version field when persisting (for example app_schema_version)
  • validate key fields after import (memo shape, context_window length)
  • never blindly trust external input; degrade gracefully on import failures