Context & continuity
History, memory, and derived context each have their own job.
The canonical transcript
The persisted conversation records authentic user, assistant, and tool lineage. Streaming views and summaries are derived representations. They do not replace the authoritative record of what was said or done.
Working within a context window
Long conversations need bounded model input. Context maintenance can summarize, archive, and prepare derived material while preserving the distinction between authentic history and the context presented to a model.
Memory is a retrieval surface
Vector-backed memory helps retrieve relevant information. It is distinct from the original transcript and from files on an authorized node. A remembered summary and its source evidence should not be treated as interchangeable.
Owner and effective agent
A conversation’s owner controls subsystem policy. The effective agent supplies the provider-facing persona, model, tools, memory, and scope for a turn. Routing to another agent does not erase the original ownership boundary.
Separate branches, explicit return paths
Storage, emotional, and right-brain forks are specialized derived conversations used by context processing. Their outputs return through defined memory, routing, archive, or status channels. The terminology describes software processing roles, not a claim about consciousness.