Expand description
Node session registry: PTY sessions the user opened survive a node restart.
Each live PTY session is journalled (provider, provider session id, workspace, model/effort,
instance number); on node start every journalled session is spawned again with the
provider’s own resume flag, under the same instance number, so HQ shows the same chat.
Off with HATCHERY_SESSION_RESUME=0.
Structs§
Functions§
- enabled
- journalled
- Sessions journalled for restart resume: every ACP session (restored through its managed
record with ACP
session/load, else started fresh under the same number), and PTY sessions of providers with a resume flag. The journal is written at spawn and on every session-id or key change (tmp + rename), never on a stop signal, so a hard kill (Windows) keeps it. - launch_
args - Launch args: resume a known id, or pin a fresh one so it can be resumed later.
- load
- new_
session_ id - note_
account_ ref - Records that the child of
session_keycarriesaccount_refas itsM4A_SESSION_ID. - path
- refresh
- remove
- set_
session_ id - Set the provider session id of an entry that has none yet.
- set_
session_ key - Record the m4a session key an entry’s runtime got (a resume reuses the old one), so the session keeps its account and nick across node restarts.
- supported
- Providers whose CLI accepts a caller-chosen session id plus a resume flag.
- upsert