applyCanvasPatch
ChatGPTApply multiple canvas mutations atomically (all succeed or none). Prefer this for builds or extracting focused supporting documents: combine new documents, source edits, and labeled edges in one patch. Position-only moves use object.update with only x/y. Empty mixed-patch content/metadata keys are omitted, not cleared; use document.update or updateObjectMetadata to clear. Pass baseSeq (v3) or expectedRevision (v2) from last read rev. Read data.staleRead and data.applied[].dropped; dropped geometry is not success of intent. Content uses Markdown. Ops: object.create (tempId for later op references), object.update, document.update, object.delete, connection.create (flow/dependency/hierarchy/cycle/overview-to-supporting-data-or-formulas; not theme alone), connection.delete, canvasDocument.update, layer.moveObjects. Markdown links need real object IDs, not tempIds. Use granular tools for simple single changes.
createConnection
ChatGPTCreate a labeled structural arrow: flow, dependency, hierarchy, cycle, or overview-to-supporting-document relation (data, formulas, specification). Overview-to-detail points to the supporting part; depends-on points to the prerequisite. Shared theme, similarity, category, and co-mention alone do not justify an edge. Read listConnections before extending an existing graph.
createLayer
ChatGPTCreate a new layer in the active project and set it as active
createMultipleObjects
ChatGPTCreate multiple focused documents at once: separate overview, data, and calculations when independently useful; keep repeated records in a table. Use clear Markdown and labeled structural edges (flow, dependency, hierarchy, overview-to-data/formulas/specification), not shared theme or co-mentions. Keep branches compact and aligned. Prefer applyCanvasPatch when combining creation with edits to existing documents.
createObject
ChatGPTCreate a single focused object document in the active project. Use type="emoji" for emoji objects, type="textLabel" only with emoji="🔤". Place inside bounds (x: 450-2550, y: 375-2125), ~200px inward from edges; aligned rows/columns are valid. Inspect occupied positions when extending a structure. Emoji are square; scale is a whole step 1–4, default 1. Meaningful content should use clear, purposeful Markdown. Separate overview, data tables, and calculations when they serve distinct reading purposes within the requested scope.
createProject
ChatGPTCreate a new project for the current user and optionally set it as active
deleteConnection
ChatGPTDelete an existing connection by connectionId
deleteObject
ChatGPTDelete an object and its related connections from the active project
getCurrentProject
ChatGPTGet metadata and summary for the currently authorized project
listConnections
ChatGPTRead graph relationships without opening documents. objectId returns both incoming and outgoing edges; fromObjectId filters outgoing, toObjectId filters incoming, and combined filters intersect. Returns edge IDs, endpoint IDs/names/layers, hasArrow, and labels. Follow endpoint IDs to traverse a branch; hasArrow=false is undirected. Default 200 results, max 500 per page; pass offset=nextOffset while hasMore=true. Reads span layers and include rev.
listObjects
ChatGPTList objects in the active project with id, emoji, name, type, x, y, layerId, and metadata. Optionally filter by layer. Default 200 results, max 500 per page; pass offset=nextOffset while hasMore=true. Metadata defaults to slim (role, category, tags only); pass includeFullMetadata=true for keywords, description, and custom fields. Use listConnections to inspect relationships.
listProjects
ChatGPTList all projects available to the current user
moveObjects
ChatGPTMove objects to new x/y coordinates. Position-only: metadata, document content, emoji, name, type, and geometry are preserved. Prefer this over updateObjects for relocate tasks. Pass one entry per object in { moves: [{ objectId, x, y }] }.
moveObjectsToLayer
ChatGPTMove objects to a specific layer by layer name or ID
openLevelry
ChatGPTOpen the Levelry canvas editor. To open a specific project, resolve its ID with listProjects and pass projectId. Omit projectId only when the user has not selected a project.
readDocument
ChatGPTRead document content for a single object by objectId
readDocumentExcerpt
ChatGPTRead a bounded excerpt from an object document by character range, or around a query match. Prefer this over full document reads when only local context is needed.
readDocuments
ChatGPTRead document content for multiple objects at once (max 20)
renewLevelrySession
ChatGPTRenew the editor session.
searchDocumentOccurrences
ChatGPTFind exact/token occurrences across object names, document content, tags, keywords, and metadata (including custom fields). Returns match snippets, counts, and pagination for grep-like discovery without reading full documents.
searchDocuments
ChatGPTToken-efficient full-text search across object names, document content, tags, keywords, and metadata. Default limit 10 (max 100). Returns one snippet per matched object plus totalObjectsMatched and hasMore. No offset cursor: if hasMore, narrow the query or raise limit (max 100). Use searchDocumentOccurrences for grep-style paging with offset. By default omits metadata and uses short snippets (compact=true). Set includeMetadata=true for slim metadata (role, category, tags; plus the matched field if needed) or includeMetadata=true and compact=false for full metadata and longer snippets.
switchProject
ChatGPTSelect a project by ID; use this ID for subsequent calls.
updateCanvasDocument
ChatGPTUpdate the project-level Canvas document (README). Not an object — do not use updateDocument. Pass the full revised content. Pass an empty string to clear it. Use this when the focused document is the Canvas document or the user asks to edit the canvas/project document.
updateDocument
ChatGPTUpdate or create visible Markdown document content for an object in the active project. Read the current document first unless you are replacing it wholesale (new structure); skipping the read risks clobbering concurrent edits or the existing structure. Preserve its intentional style unless the user asks to reformat it. Pass an empty string to clear the document body. Do not use this for structured metadata or special fields; use updateObjectMetadata. REQUIRED: if this object's metadata is empty (no tags/keywords/role/category), also call updateObjectMetadata in the same turn with a few short tags (optional role) derived from the content — never for RULES/MEMORY or non-empty metadata, never dump a Metadata section into the document body, and do not set description/keywords.
updateObjectMetadata
ChatGPTUpdate structured metadata for an object (key-merge: omitted keys are kept). Prefer small tags (1–5 short labels) plus optional short role; optional category only when useful. Pass empty tags/keywords arrays or empty role/category/description to clear those fields. An empty metadata object is a no-op — it does not clear all fields. Do not set description (prose belongs in the document body). Do not set keywords (use tags). Do not restate the document into metadata. Custom fields remain allowed when the user asks. Call this in the same turn after updateDocument when the object had empty metadata (lazy enrichment). Skip RULES/MEMORY and objects that already have non-empty metadata.
updateObjects
ChatGPTPatch one or a batch of objects (position, scale, emoji, name, metadata, Markdown document). Only provided fields change; omitted fields are kept. Empty or whitespace content and empty metadata keys (tags:[], role:"", …) are ignored and do not clear existing values. To clear a document, call updateDocument with empty content. To clear metadata fields, call updateObjectMetadata with empty tags/keywords arrays or empty role/category/description. Prefer moveObjects for position-only relocates. layerId is not accepted here — use moveObjectsToLayer. Name and emoji cannot be cleared (min length 1). Preserve intentional document structure when you do pass content.