Remove the proactive-memory preference and its cross-platform runtime plumbing so memory behavior has one clear contract. Normalize away the obsolete setting and keep source type limited to provenance.
Share entities resolved during one Agent run so natural-language subject changes can safely use memory. Validate entity memory scope before permanent deletion to prevent stale IDs from deleting another subject's memory.
Include the normalized time-filter request in structural-inspection fingerprints while retaining the cursor's resolved absolute range, so stable relative pagination works but changed filters are rejected.
Interpret recent-day filters as complete local calendar days and persist the first resolved range in structural-inspection cursors so pagination remains stable across clock and date changes.
Resolve private ranking display names, apply privacy preprocessing consistently across cached recaps and structural labels, and bound model-visible overview payloads to the configured tool budget.
Keep shared-interaction evidence, pagination, and token truncation accurate while bounding anchor inputs, proximity scans, contact discovery, and reply aggregation. Resolve participants across all imported history and isolate damaged database candidates without changing the final analysis tree.
Use the all-history contact snapshot and validate raw session candidates inside the cancellable scan loop. Bound model-visible structural results to the configured token budget while preserving full tool details and continuation metadata.
Keep full entity resolution details available to the application while bounding the model-visible payload with the runtime tokenizer. Preserve coverage and truncation metadata, prioritize usable scopes fairly, and cap ambiguity hints to avoid duplicate unbounded source lists.
Count complete tool payloads with the runtime tokenizer so Chinese evidence respects the configured model context budget.\n\nResolve contact sources incrementally with cancellation checks and targeted member lookups to keep long cross-chat investigations responsive.
Make unscoped candidate preparation honor cancellation and wall-time limits before message scanning. Enumerate database files lazily, retain only the newest eligible sessions, and resolve owners without loading full member statistics.
Use the all-history contacts snapshot when translating stable contact references into cross-chat scopes, so older source sessions are not silently excluded from historical analysis.
Replace message content and structured user entity references in one database statement. Treat omitted references as an explicit clear so edited prompts cannot replay stale identities.
Prevent generic assistant or summary writes from persisting structured entity references. Keep compressed-summary references on the dedicated internal path and ignore malformed assistant references during replay and compression.
Render circular member avatars on ranking lists, person-based charts,
hot-repeat originators, and catchphrase rows. Look up avatars from the
member table instead of bloating ranking queries, and clip ECharts
photos so they match the HTML lists.
Keep the shared runtime in tool_running until every concurrent call finishes. Route renderer progress and completion by toolCallId so one result cannot hide another active tool.
digua