Search the user's unified memory across all AIs. Use this at the start of a task to recover context. source filters to one AI (chatgpt/claude/gemini); type filters by fact/preference/project/relationship/conversation/note. semantic_weight (0.0–1.0) tunes hybrid search: 0.0 = pure keyword, 1.0 = pure semantic, 0.5 = equal. Use -1.0 (default) for the config default. Results are limited to the memory scopes this client is allowed to see.

List the user's EleSync projects and their short ids. Use this before recall_project when the user refers to a project by name.

Search only the chats and memories collected in one EleSync project. Use the id from list_projects. This is safer and more relevant than a whole-vault recall for project-specific work.

Return the approved project knowledge sheets, tasks, decisions, handoffs, and collected memories as a compact context package.

Create a new durable document inside one EleSync project.

This is a mutating tool: it writes the title and body to local vault storage and updates the project manifest. It does not send content to an AI provider, and it does not overwrite an existing document; each call creates a new document. project_id must identify an existing project and kind must be one of knowledge, task, decision, or handoff. Call it only after the user has approved saving the content. A project-bound connection can use this only for its assigned project and only when it has project write permission.

Save a new memory to the user's vault so every other AI can see it too. Use for durable facts, preferences, project state — not throwaway chatter. Pass your identity in source (e.g. 'claude', 'chatgpt', 'gemini') so provenance and conflict detection work correctly across AIs. scopes (comma-separated, e.g. "legal" or "writing,coding") puts the memory in a compartment only clients granted that scope can read; omit it to make the memory general (visible to all) unless this client has a default scope. supersedes (a memory id or 8-char prefix) marks that this memory replaces an older one when a fact changes rather than adds (a move, a rename, an updated preference): the old memory is kept as history but hidden from recall, and history can then show what changed and when. template names a structure from the memory://templates resource and applies its type/tags/scopes defaults; fields is a JSON object filling that template's placeholders, e.g. {"who": "Ana", "date": "2026-07-25"}, which builds the content for you so repeated captures stay comparable.

Delete a memory by id (full or 8-char prefix shown by recall).

Summary of what's in the vault: totals by source and type (limited to the memories this client's scope can see).

List the memory scope compartments this client can see, with how many memories each holds. general is the default, everyone-can-see bucket.

Surface facts in the vault that look like they're about the same thing but disagree (e.g. two different stated locations), each with its source and date. Use this to spot stale/contradictory memory and confirm with the user.

Show the supersession history of a memory (id or 8-char prefix): every version that led to or replaced it, oldest first, with when each changed. This is the "what was true when" view — use it to see what a fact used to be after a memory was superseded (via remember(..., supersedes=...) or by resolving a conflict).

View the audit trail of operations on this vault. Shows who did what and when — useful for reviewing which AI clients are reading or writing memories. Audit logging is enabled by default; set ELESYNC_AUDIT=0 to disable it. last limits the number of entries (newest first). operation filters to a specific op (REMEMBER, RECALL, FORGET, SCOPE, STATUS). client filters to a specific client label.