MemoryRouter

Use this to complete a consolidation started with consolidate_memories. Submits the reflection entries you wrote for the checked-out batch. Each entry is standalone prose with an importance rating from 1 to 10. The server stores the reflections, links them to their source memories, and marks the sources consolidated. Safe to retry: committing the same batch twice returns the original result.

Use this when the user asks to consolidate, reflect on, or organize their memories, or when memory_status or a search result reports that consolidation is available. Checks out a batch of unconsolidated memories from the connected vault under a 15 minute lease and returns their texts plus server instructions for writing reflection entries. Follow the returned instructions to write the entries, then call commit_reflections with the batch_id. Nothing is modified until commit.

Use this to retrieve memories from a specific time window, for questions like 'what happened last week' or 'most important things this month or lately'. Tier guide: tier 3 holds the highest-level consolidated reflections (identity, principles, what matters most), tier 2 holds reflections of specific meaningful events, tier 1 holds raw verbatim memories. Match filters to the question. Time-bounded 'most important things last month' questions: tiers [3], importance 8 or higher, generous max_tokens. Big-picture overviews of a period like 'what do you know about July': tiers [3] with no importance filter. Specific lookups in a window like 'find the decision about X': omit tiers to include all tiers and leave importance unset; a high importance filter hides the specific detail you are looking for. Specific meaningful events without raw noise: tiers [2] or [2, 3]. Exact verbatim detail: if the detail is not already in context, drill down by lineage: search tiers [3] to find the relevant thread, then use inspect_memory on the result to see the tier 2 reflections it consolidated, then inspect those to reach the raw memories underneath. Direct tiers [1] search works when you know exact wording to match. Mixing tiers is legitimate; adjust the importance threshold to the question rather than always setting it high. Only reflections (tiers 2 and 3) carry importance ratings. Omit query for a chronological review; include query to rank by relevance within the window. Resolve relative phrases like 'lately' to concrete ISO dates before calling. Reads only the OAuth-selected vault and does not modify it.

Use this only when the user explicitly asks to delete specific memories and confirms after seeing what will be deleted. Permanently deletes the identified memories from the connected vault by their ids from search or inspect results. Deleting a raw memory does not delete reflections built from it. This cannot be undone. For deleting everything, forget_all_memories is the separate whole-vault tool.

Use this only after the user explicitly requests a full-vault deletion, receives a permanent whole-vault warning, and provides the exact confirmation phrase in a later turn. Permanently deletes every memory in the OAuth-selected vault. This cannot delete one memory; use delete_memories for specific items. Requires separate memories:delete authorization and destructive host confirmation.

Use this when exact details or provenance matter: when the user asks why the assistant believes something, or wants to see the source memories behind a reflection. Given a memory id from a search result, returns the lower-tier memories that reflection was built from, or, for a raw memory, which reflections it was consolidated into. Use repeatedly to drill down from a high-level reflection to the tier 2 reflections and raw memories beneath it. Read only.

Use this when the user asks whether MemoryRouter is connected, which opaque vault binding is active, or how many memories/tokens the connected vault contains. Does not reveal the memory key.

Use this legacy compatibility alias when a client explicitly requests a tool named search. It searches the OAuth-selected MemoryRouter vault without modifying it. For normal ChatGPT recall, prefer search_memories; this tool does not provide citable item permalinks or Company Knowledge fetch support.

Use this before responding to each new user message, to retrieve context relevant to the message and ongoing conversation, even when the user does not explicitly ask you to remember. Form a concise semantic query using the people, projects, preferences, goals, decisions, or events implied by the message; for follow-ups, include enough recent conversational context to make the query meaningful. Search all tiers (omit tiers and importance) unless there is a reason to narrow. If a MemoryRouter recall block was already supplied for this message, that search already ran; search again only for more detail. Read only: searches the user's OAuth-selected vault. Tier guide: tier 3 holds the highest-level consolidated reflections (identity, principles, what matters most), tier 2 holds reflections of specific meaningful events, tier 1 holds raw verbatim memories. Match filters to the question. Broad 'most important things we have done' questions: tiers [3], importance 8 or higher, large limit such as 250. Big-picture overviews like 'what do you know about this area or period': tiers [3] with no importance filter. Specific lookups like 'find the decision about X': omit tiers to search all tiers blended and leave importance unset; a high importance filter hides the specific detail you are looking for. Specific meaningful events without raw noise: tiers [2] or [2, 3]. Exact verbatim detail: if the detail is not already in context, drill down by lineage: search tiers [3] to find the relevant thread, then use inspect_memory on the result to see the tier 2 reflections it consolidated, then inspect those to reach the raw memories underneath. Direct tiers [1] search works when you know exact wording to match. Mixing tiers is legitimate; adjust the importance threshold to the question rather than always setting it high. Only reflections (tiers 2 and 3) carry importance ratings. For time-window questions like 'what happened last week' or 'lately' use date_search_memories instead.

Use this after each user message that contains anything worth knowing later, even if the user does not say 'remember this'. Default to saving: a missed memory costs far more than a redundant one, so when in doubt, save. Each of these is a trigger on its own: the user makes a decision, states a preference, gives a fact about a person, project, company, or tool, sets a plan or commitment, makes a correction, or reports progress or an outcome. Save one concise, self-contained memory per call, written so it makes sense on its own later, with the exact wording, numbers, dates, names, and identifiers the user gave; several facts in one message mean several calls. Record what the user actually said, not your inference. There is no update tool: to supersede outdated information, save the corrected fact and say what it replaces. Never save passwords, API keys, tokens or other credentials, payment card numbers, government ID numbers, or protected health information, and never save whole transcripts. Writes one memory to the user's OAuth-selected vault. If the user asks what was saved, answer truthfully from the tool result.