Session discovery: conversations become unreachable, plus ghost and duplicate project entries

The issue of conversations becoming unreachable and duplicate or ghost project entries in the major agent CLIs is a prevalent problem, as detailed in a comprehensive report titled "Session discovery: conversations become unreachable, plus ghost and duplicate project entries." The reporter, who works with roughly ten active conversations, highlights that the issue lies not in the model's performance but in the inability to address the conversations due to their unaddressable status, some of which cannot be opened at all.

Additionally, the project list has grown to include ghost entries and duplicates, creating confusion over where to find relevant information. Two reports from different angles describe the same class of failure. The first report suggests that the CLI should suggest starting a new conversation when resuming a large or stale session, as resuming can drag in outdated context.

The second report asks for conversation-scoped visibility, allowing an agent to recall sibling threads within a channel without reaching into other channels. These issues stem from storage and addressing problems that underlie the entire system. The report rules out prompt rewriting and larger context windows as solutions, as they do not address the fundamental problem of missing addressable storage.

A new layer of memory is required to write facts down with source and timestamps, giving each one an address for retrieval from different sessions. Four rules are enforced in HyperMarrow, a local-first memory layer for agents on Windows, to ensure durable storage, retention policies, recall quality, and bounded forgetting. These rules address the reported symptoms, ensuring conversations remain addressable, eliminating ghost entries, and providing conversation-scoped visibility.

However, challenges remain in hybrid ranking, short queries against long records, and session identifiers, which can affect memory layer performance. Despite these limitations, HyperMarrow aims to provide a more reliable solution for agents across multiple sessions.