add_annotation
ChatGPT[Vault] Write an annotation to a source — a real-time extraction-style note by the source's own author while they are working on it. Annotations are the author's own voice, distinct from seeds (Neuron extraction) and marginalia (later-mind marks). Use when the source's author wants to mark their own decisions, insights, or notes during authoring.
add_marginalia
ChatGPT[Vault] Write marginalia to an existing source — a correction, tension flag, evolution narrative, or commentary by a later mind. Marginalia is distinct from seeds (Neuron extraction) and annotations (author's own marks). Every marginalia records its scribe identity and markType for cross-AI provenance. Use when correcting prior content, flagging contradictions, narrating concept evolution, or adding enriching commentary.
add_sources_to_folder
ChatGPTAdd one or more sources into a manual Vault folder.
batch_status
ChatGPTCheck the progress of a batch extraction job.
compose_workspace
ChatGPT[Workspace] Create a fully configured workspace with Container, Studios, Lanes, and governance seeds. This is the 'talk it into being' operation — describe what you need, and the workspace appears. Touches workspace primitives only; the vault is unaffected. Use when the user wants to create a project board, organize a goal, structure a client/project/life domain, turn a messy plan into lanes/cards, or talk a workspace into existence. Returns containerUrl + per-studio studioUrl for direct deeplinks.
create_card
ChatGPT[Workspace] Drop a card into a lane. Cards are the workspace coordination surface (tasks, notes, references, handoff packets). Two types: 'note' for workspace-native content, 'reference' for pointers at a vault source or seed. Pass laneId, or resolve by name: laneName + (studioId or studioName). Use when the user wants to capture a task, note, idea, or pointer-to-source in a specific lane; this is the most common writing operation in workspace mode. Returns laneName, studioName, containerName, and a studio deeplink (cardUrl) so the user can open it.
create_container
ChatGPT[Workspace] Create a new top-level container (Mission Horizon). Containers organize studios into broad operational domains. Use when the user wants a new project domain at the top level; for a fully configured workspace with studios + lanes in one call, prefer compose_workspace.
create_default_studio_lanes
ChatGPT[Workspace] Create standard workflow lanes (Intake, Doing, Needs Context, Review, Shipped) for a studio workspace. Idempotent — will not duplicate existing lanes.
create_lane
ChatGPT[Workspace] Add a lane to a studio's deck. Lanes are columns that hold cards. Pass either studioId or studioName. Use when the user wants a new column/swimlane inside an existing studio (e.g., 'add a Blocked lane', 'add a Calendar lane'). For a fresh studio with all lanes specified, set them via compose_workspace or create_studio instead.
create_skill
ChatGPTCreate a new skill. Skills define WHAT to extract from specific content types — custom fields, scoring criteria, domain-specific analysis. Aero writes these conversationally. Default skillType is 'extraction' (domain packs that shape how sources become seeds); pick 'evaluation' for Archivist brief evaluators, or 'production' for Assemblist content pipelines.
create_source
ChatGPT[Vault] Save a conversation, document, or transcript to the vault. Sources always live in the vault — they are never owned by containers. Automatically extracts decisions, frameworks, key passages, and more with full provenance. Use when the user says 'save this conversation,' 'send this to my vault,' 'remember this,' or 'import this document.'
create_studio
ChatGPT[Workspace] Create a studio workspace within a container. Studios are configurable surfaces with an ordered set of modules (source, writer, deck, brief, production, etc.), pinned sources, and lanes for organizing cards. Pass either containerId or containerName. Defaults to [source, writer, deck] when modules omitted. Use when the user wants to add a new working surface inside an existing container (e.g., 'add a research studio to my Finance container'); for a brand-new container with studios from scratch, prefer compose_workspace.
create_vault_folder
ChatGPTCreate a new manual collection folder or dynamic smart search folder in the user's Vault.
delete_card
ChatGPTDelete a card from its lane. Soft-delete by default (recoverable); pass hard=true for a true DB delete. The vault is unaffected — even reference cards just remove the pointer.
delete_container
ChatGPTDelete a container. Studios and lanes are deleted, but seeds are orphaned (kept in vault). Requires confirm=true.
delete_lane
ChatGPTDelete a lane. Seeds in the lane are moved to the studio's first remaining lane before deletion.
delete_skill
ChatGPTDelete a custom extraction skill. Seeds already extracted with this skill are not deleted.
delete_source
ChatGPT[Vault] Delete a source exclusively claimed in the active Realm. Shared sources require a separate shared-write policy.
delete_studio
ChatGPTDelete a studio and its lanes. Seeds are orphaned. Requires confirm=true.
delete_vault_folder
ChatGPTDelete a Vault folder or smart collection. Underlying sources remain safe in your vault.
get_brief_status
ChatGPTPoll a research brief's status. Accepts sourceId (preferred — what request_brief returns) or briefId (legacy — for briefs that pre-date the brief-as-source collapse). Status values: 'synthesizing' (brief being assembled, no content yet), 'extracting' (legacy in-flight status), 'complete' (brief ready — retrieve via get_source), 'error' (synthesis failed; details in metadata.error).
get_capabilities
ChatGPTIntrospect the current MCP session and orient yourself to the vault surface. Returns: which canonical tools you can actually call right now (grouped by category), the canonical primitives and how they relate, and the typical-flow navigation spine. Call at session start to preflight what's permitted AND to load the right mental model. Pair with get_vault_summary for a snapshot of what's in the vault.
get_card
ChatGPTRead a single card's full content, metadata, and location (lane, studio, container). Use when GPT or the user needs the full body of a specific card by ID.
get_container
ChatGPTGet a container's full structure — studios with their lanes, framework seeds, routing rules, and optionally recent seeds. This is how the morning brief reads each desk's state.
get_container_stats
ChatGPTQuick stats for a container — studio count, seed count per studio, lane counts, category breakdown. Designed for morning brief at-a-glance summaries.
get_container_structure
ChatGPT[Workspace] Read a container's full organizational structure — all studios, lanes, card counts. Pass either containerId or containerName. Note: This is a deep structural tree inspector. To list top-level domains, prefer list_containers.
get_context
ChatGPTGet a scoped context snapshot from the vault — recent seeds, top tags, and category stats. Use this to understand what the user is working on.
get_current_realm
ChatGPT[Realm] Get the currently active Realm context for this AI session. Operations against the vault (sources, seeds, extractions) are partitioned to this realm.
get_deck_state
ChatGPTRead the full deck state for a studio — all lanes with their cards. Use when you need to know what work is active. Pass either studioId or studioName.
get_desk_state
ChatGPTRead the full workstation desk orientation for a studio: assigned resident agent, equipped 2-4 tool bouquet, 5-lane agent runway deck status, and active studio context brief. Pass either studioId or studioName.
get_excerpt
ChatGPTRead a bounded page of one original Excerpt and verify its exact UTF-16, end-exclusive Source coordinates and revision. Changed or inaccessible originals fail explicitly. Reading creates no Seeds, model calls or Source writes.
get_lane_contents
ChatGPTGet all seeds in a specific lane. Faster than filtered search for reading a single lane's contents — e.g., 'show me everything in the Qualified lane'.
get_skill
ChatGPTRead one skill's full extraction recipe — instructions, schema with field names, and trigger conditions. The schema[].field values are the category names you can pass to search_vault's category filter to find seeds extracted by this skill.
get_source
ChatGPTOpen one known source by id or title and read everything in it — abstract, all seeds grouped by their 8 canonical categories and any skill-defined categories, and assets. Use this when you already have a specific source in mind. Not for searching — use search_vault to find seeds across the vault, or search_sources to locate which source you want first. By default the response includes raw content; for sources larger than ~100k characters the content is auto-omitted with omittedReason set, in which case re-request with explicit include: ['content'] if you really need it.
get_vault_gaps
ChatGPTAnalyze the vault's coverage — categories with low density, topics mentioned but unexplored, unextracted sources. Use this to decide what to explore next.
get_vault_summary
ChatGPTReturns a high-level snapshot of the entire vault: source count, seed count, category distribution, recent sources, and topic clusters. Use at session start or when you need to understand the full knowledge landscape.
graduate_card
ChatGPT[Workspace → Vault] Graduate a workspace card into a canonical vault Source. This is the bridge across primitives: cards are ephemeral workspace coordination objects; sources are durable vault ground truth. Graduation creates a Source from the card's content (derivationType='conversation_graduated', parent linkage preserved) and optionally triggers extraction. Use when a card has earned permanence — its content would create friction in a future session if it stayed inside the workspace and got deleted with the lane.
list_cards
ChatGPT[Workspace] List cards within a studio or specific lane. Pass studioId/studioName or laneId/laneName. Supports optional status filter and limit.
list_containers
ChatGPT[Workspace] List top-level containers (Mission Horizons). Scoped by default to the active Realm to prevent cross-domain leakage. Pass optional realmId to inspect a specific realm, or set allRealms: true for an account-wide overview. Each entity in the response includes a ref field (the human-readable name) and an id field. When writing prose for the user, reference entities by their ref. Use id only when making subsequent tool calls. Never echo raw UUIDs into conversational text.
list_lanes
ChatGPTList all lanes in a studio with card counts and optional preview (first 3 card titles). Pass either studioId or studioName.
list_local_models
ChatGPTList available local language models, parameter sizes, and quantization levels from the configured inference runtime.
list_pins
ChatGPTList sources currently pinned to a studio with seed counts. Returns metadata + abstract by default — call get_source on a specific pin when you need full content.
list_production_skills
ChatGPTList available Production Skills (Assemblist work orders) grouped by domain. Production Skills produce deliverables from vault knowledge — articles, briefs, proposals, voice captures, and more. Use to discover what can be produced, then invoke with produce_deliverable.
list_realms
ChatGPT[Realm] List all Realms available to your account (e.g. Multiplist, Gaia Futures, Client Workspaces). Returns realm id, name, description, and default status.
list_skills
ChatGPTBrowse available extraction skills — domain-specific recipes that define what to extract beyond the canonical 8 categories. Returns skill name, description, domain, type, and field count. To see a skill's actual extraction fields (which are the category names usable in search_vault's category filter), follow up with get_skill.
list_source_assets
ChatGPTList detected assets (work products, drafts, code blocks, tables) found in a source prebiotically. Returns detection suggestions with exact character ranges, lines, types, confidence, and formatted snippets. Use rescue_excerpt to promote confirmed assets into sovereign sub-sources.
list_studios
ChatGPT[Workspace] List Studios within a container or across the active Realm. Pass containerId or containerName to filter to a specific container.
match_skills
ChatGPTFind which extraction skills apply to a given source based on content and trigger conditions. Use when you've imported a source and want to know which skills would produce the most relevant custom extraction alongside the canonical 8 categories. Only returns extraction-type skills — evaluation and production skills are routed through Archivist and Assemblist respectively.
move_card
ChatGPTMove a card to a different lane. Lane can be specified by ID or by name within the card's current studio. Use this for natural commands like 'move this card to Shipped'.
move_source_to_realm
ChatGPT[Realm] Transfer a source exclusively to a different Realm. Unclaims from source realm and claims into target realm.
pin_source
ChatGPT[Workspace] Pin a source to a studio so its seeds load into foreground context. Creates a workspace-side join row; the source itself is unmutated. Pins are an uncapped, ephemeral working set. Use when the user wants to bring a vault source into a studio's working set (e.g., 'pin the strategy doc to Sprint Planning'); the source is not moved — only referenced.
produce_deliverable
ChatGPTInvoke a Production Skill to generate a deliverable from vault knowledge. This is the Assemblist engine — it takes a skill (work order), vault sources for context, and optionally a voice pack, then produces structured output (articles, briefs, proposals, etc.). The output is auto-saved to the Outbox. Use after list_production_skills to discover available skills.
query_agent_log
ChatGPTRead external-agent telemetry — what agents (Greenroom, BD, Funding Scout, etc.) have been doing in your vault. Filter by agentIdentifier, container, workflow, outcome, or date range. Each entry records the agent's plan, actions, outcome, and cost; rows flagged with promotion signals (shouldPromoteToMemory, shouldNotifyHuman) are review-worthy. Distinct from audit_log (low-level MCP audit) — this is operational visibility for agent oversight.
query_local_model
ChatGPTExecute inference on a local language model with specified prompt, system prompt, and temperature parameters.
remove_source_from_folder
ChatGPTRemove a source from a manual Vault folder. (The underlying source remains intact in the Vault).
rename_vault_folder
ChatGPTRename an existing Vault folder or smart collection.
request_brief
ChatGPTSynthesize a research brief across vault sources on a topic. Two modes: mode: 'full' (default) runs async — Multiplist synthesizes a research document and saves it as a vault source; returns immediately with a sourceId, poll via get_brief_status until complete, then read via get_source. mode: 'retrieval_only' returns synchronously with ranked seeds + constellation links — no LLM call on Multiplist's backend, no vault write. Use retrieval_only when you want Archivist-quality retrieval to synthesize yourself. After synthesizing from retrieval_only results, offer the user: 'Want me to save this research to your vault?' If yes, call create_source with your synthesized response (sourceType='archivist_brief', tags=['archivist-synthesis', 'retrieval-capture']).
rescue_excerpt
ChatGPTLift a detected draft, code block, or substantial content chunk from a source into its own sub-source. The excerpt becomes a sovereign vault primitive with verbatim content and provenance linking back to its parent source. Use when a source contains work products worth rescuing — guidebook entries, spec sections, email drafts, code modules.
resolve_workspace
ChatGPT[DEPRECATED: Prefer list_containers, list_studios, or passing human-readable names directly into entity tools] Translate human-readable workspace names to system IDs.
search_excerpts
ChatGPTHERO ANCHOR (Surface Before Compute): Find bounded, deterministic original passages across Sources at $0 LLM compute. Excerpts are coordinate-bounded windows (300B–1.5KB) split along natural speaker turns and genres (code, drafts, specs). Returns exact Source coordinates, revisions, and previews. Use get_excerpt to verify and read an intact original passage before quoting it. Search and reading create no Seeds or model calls.
search_sources
ChatGPTFind sources (whole conversations/documents) by title, date, type, or keyword — the card catalog of your vault. Use this when you want to locate documents, not the meaning inside them. Not for retrieving extracted seeds — use search_vault to query seeds across sources, or get_source once you know which source you want. COLUMN COVERAGE: Queries match against the source title + rawContent (the full imported body) + abstract. A term that only appears inside a doc's body — not its title or abstract — will still surface here. (Contrast with search_vault, which matches seeds against title/content/summary and matches sources against title/abstract only — if you want body-content matching on sources, reach for search_sources.) Note: rawContent search uses ILIKE backed by trigram indexes (not full-text search) to avoid tsvector overflow on very large sources. Supports the same inline query syntax as search_vault (delimiter-insensitive, "quoted phrases", -exclusions, field operators like title:, tag:, after:, before:, last:). Fuzzy correction IS applied — same as search_vault and the UI Vault search — so common typos are silently corrected against your vault vocabulary. Multi-word queries also get the same automatic relaxation: if the strict all-terms match finds nothing, the engine retries with a per-token OR ranked by term coverage and sets meta.relaxed: true. ZERO-RESULT RECOVERY: When the query still returns zero results (even after relaxation), the response includes meta.recoveryHints — short hints for retrying with a different shape (fewer terms, drop quotes, remove connector words). Use these to self-correct rather than reporting a hard miss. Each entity in the response includes a ref field (the human-readable name) and an id field. When writing prose for the user, reference entities by their ref. Use id only when making subsequent tool calls. Never echo raw UUIDs into conversational text.
search_vault
ChatGPTFind seeds across your entire vault by semantic query. Use this when you want extracted meaning — decisions, frameworks, golden passages, or skill-defined categories like "Product Moves" — regardless of which source they came from. The category filter accepts the 8 canonical categories AND any skill-defined category name; it matches both the stored form (product_moves) and the human-readable form (Product Moves). Not for finding whole sources — use search_sources for that, or get_source to open one you already know. WHAT YOU GET BACK: 1. results — seeds matching the query, scanned against seed title + content + summary + tags. This is the primary list. 2. sourceMatches — sources whose title OR abstract matches, returned alongside as a fallback for queries that landed in this tool but were really shaped for the catalog. Capped small. 3. primaryHit (when emitted) — a single dominant pointer (e.g. an exact title match on a source). Always follow this first when present. QUERY SYNTAX: - Delimiter-insensitive: underscores, hyphens, spaces, and dots are interchangeable (so "THE_GRAIN_OF_TOOLS" and "the grain of tools" return the same results). - "quoted phrases" are matched as exact substrings. - -word excludes that term from the results. - Field operators (all optional, combine freely): - title:foo / title:"my exact title" - tag:strategy - category:decision | framework | golden_passage | definition | action | question | offer | emergence - type:seed | annotation | marginalia - mark:correction | tension | evolution | commentary - scribe:claude-opus-4-7 - after:2026-03-01 / before:2026-04-15 / last:7d (also 24h, 2w, 3m) Structured parameters (category, temperature, etc.) take precedence over parsed field operators when both are supplied. PAYLOAD BEHAVIOR: By default each hit returns a lean seed projection — the fields you need to read, quote, and navigate (id, ref, title, content, summary, tags, extractionCategory, subCategory, skillId, confidence, durability, pinned, primitiveType, markType, scribe, sourceId, sourceType, timestamps). Heavy/internal columns (embedding, metadata, lineage, intentVector, nucleus, positional offsets, decayWeight, vestigial workspace ids) are omitted so broad queries stay inside the tool-result token limit. content and summary are also truncated to ~280 chars with meta.truncated: true set. For the complete text or positional provenance of any hit, call get_source on its sourceId. If you must have the full seed row inline (all columns, untruncated), pass verbose: true — use it sparingly. FUZZY CORRECTION: Query tokens are auto-corrected against the user's own vocabulary (source titles, container names, skill names, product terms). Talk-to-text mis-hearings like "Multiplus" → "Multiplist" or "arrow" → "Aero" silently resolve to the right term. When a correction is applied, the response includes a meta.corrections array so you can note in your reply what the query was rewritten to. When a match is only phonetic, it comes back in meta.suggestions as a "did you mean" candidate rather than being applied silently. Phonetic suggestions are only emitted for short (≤7 char), all-lowercase, length-close tokens — longer or proper-noun-shaped inputs are treated as deliberate. MULTI-WORD RECALL: Natural multi-word queries are handled automatically. The engine first tries a strict match (all terms together); if that finds nothing, it transparently retries with a per-token OR and ranks by how many of your terms each hit contains. So "decommissioning liability bonding gap Gulf of Mexico" surfaces the same strong seeds that "decommissioning" alone would, best-covered first — you don't need to manually decompose a query into single terms. When this relaxation fires, meta.relaxed: true is set. ZERO-RESULT RECOVERY: When the query still returns zero results (even after relaxation), the response includes meta.recoveryHints — short hints for retrying with a different shape (fewer terms, drop quotes, remove connector words). Use these to self-correct rather than reporting a hard miss. Ea…
share_source_to_realm
ChatGPT[Realm] Share an existing extracted source to another Realm ($0 re-extraction cost). Adds a realm pointer so all extracted seeds and provenance instantly appear in the target realm.
switch_realm
ChatGPT[Realm] Switch your active Realm operating context. Future vault reads and writes will operate in the target realm.
unpin_source
ChatGPT[Workspace] Remove a source from a studio's pinned sources. Removes the workspace-side join row only; the source itself is unaffected. Use when the user is done referencing a source in a particular studio (e.g., 'unpin the old draft from Active Sprints'); does not delete the source.
update_card
ChatGPTUpdate a card's title, content, status, or semantic shape (cardType, emergenceType, draftStatus). Use move_card to change which lane a card lives in. Use this for natural commands like 'rename to ...', 'mark it ready', 'tag as emergence vector'.
update_container
ChatGPTUpdate a container's name, description, or type. containerType changes are blocked once any agent is assigned (prevents orphaning agent configs); free to change while no agent is assigned.
update_lane
ChatGPTRename a lane, update its position, or modify metadata.
update_skill
ChatGPTUpdate a custom skill's name, description, domain, instructions, or skillType.
update_source
ChatGPTUpdate a saved source with new content. Supports 'append' (adds content to the end of the existing source) and 'replace' (replaces existing content). Extracts new insights from added content.
update_studio
ChatGPTUpdate a studio's name, description, or module configuration.