WindowedSummary
autogen.beta.network.views.builtin.WindowedSummary #
Keep the last recent_n visible substantive envelopes verbatim; fold everything older into a single :class:CompactionSummary at the head of the projection.
Bounds prompt size at any turn count — the projection is at most recent_n + 1 events regardless of WAL length. CompactionSummary is recognised by autogen/beta/policies/conversation.py so it renders correctly in the LLM-facing message stream.
The summary is a static stat-style line ("Earlier in this channel: N messages from a, b.") — no LLM call. Use :class:NamedWindowedSummary for N-party channels where the speaker labels matter.
Source code in autogen/beta/network/views/builtin.py
project async #
project(wal, *, participant_id, channel, render_envelope, name_for=default_name_resolver)