append_to_doc
ChatGPTAdd Markdown to an existing document without changing what is already there: at the end, or at the end of one section. Use it for follow-ups — a new finding, a step, a key fact, a dated update. Nothing is distilled; your Markdown is stored as written. The previous version is kept. To change or remove existing text, use revise_doc.
create_doc
ChatGPTWrite a document straight into DocuMemory from Markdown you compose yourself, with no distillation pass. Use this when asked to author something ('write a setup guide for Docker', 'document this module'). Use save_chat_as_doc or save_memory instead when the material is an existing discussion — those extract from a transcript, this one stores exactly what you write.
create_project
ChatGPTCreate a project to group documents under. Only needed to set a description up front or to create one before any documents exist — save_chat_as_doc, save_memory and create_doc all create a project on demand when given project_name.
delete_project
ChatGPTDelete a project. DESTRUCTIVE: with the default strategy its documents are deleted too and cannot be recovered. Confirm with the user before calling, and prefer strategy 'reassign' to keep the documents and move them to Inbox. Never call this to 'tidy up' on your own initiative.
list_doc_versions
ChatGPTShow a document's version history, newest first: who changed it, when, and the note they left. Read an earlier version with read_doc and its version number.
list_projects
ChatGPTList the projects you can access in this workspace, with your access level on each (VIEWER can read; EDITOR can also save documents; MANAGER can also change access). In an organization, call this before saving to pick a project you can write to.
list_recent_docs
ChatGPTReturn the most recent documents in this workspace. Defaults to architectural decision records (ADRs); pass doc_type to list memories or another type instead, or 'ANY' for all types.
load_project_context
ChatGPTReturn every saved MEMORY for a project as one token-optimised block, ready to inject into your working context. Call this at the START of work on a project — before proposing architecture, editing config, or writing migrations — so you are operating on this team's actual constraints rather than defaults. Cheaper and more complete than searching for rules one question at a time.
read_attachment
ChatGPTRead an attachment by its ID, including images referenced by /api/v1/attachments/<id> in a saved document. Works for files you uploaded and files linked from documents you can read. Returns image content for images and a binary resource for other files.
read_doc
ChatGPTFetch the full Markdown of one document by id, as returned by search_docs or list_recent_docs, ending with its current version number. For MEMORY documents the injectable context snippet is returned first, so it can be lifted straight into a system prompt. Pass version to read an earlier version from list_doc_versions.
restore_doc_version
ChatGPTMake an earlier version of a document current again, for example to undo a revision. The version being replaced is kept, so a restore can itself be undone. Confirm with the user before restoring over someone else's change.
revise_doc
ChatGPTReplace an existing document's body with a revised version you write, when its content is outdated or wrong or needs restructuring. Read it with read_doc first and send the complete new body — anything left out is removed from the current version. The previous version is kept and can be restored with restore_doc_version. To only add material, use append_to_doc.
save_chat_as_doc
ChatGPTDistil a discussion into a structured document (problem, decision, final code, tags) and save it to DocuMemory as a new document. Pass the relevant part of the discussion in its original wording rather than a summary. To update a document that already exists, use append_to_doc or revise_doc instead.
save_memory
ChatGPTSaves discussion excerpts, facts, or system rules as a reusable memory context document for AI re-ingestion and team handoffs. Produces standalone key facts plus an ai_context_snippet ready to paste into another model's system prompt. Use this — rather than save_chat_as_doc — when the value is standing context (architecture rules, constraints, conventions, config, domain knowledge, stated preferences) rather than a decision that was reached.
search_docs
ChatGPTHybrid semantic + keyword search across this workspace's documentation. Use it to recall prior decisions, conventions, and fixes before suggesting new ones.
upload_attachment
ChatGPTUpload actual file bytes supplied by the user. Do not invent binary content or fetch arbitrary URLs. Supply base64 bytes of a static image, PDF, UTF-8 text, ZIP, or Office file up to 3 MB. Returns a private Markdown link to include in create_doc or save_chat_as_doc. For large files, prefer POST /api/v1/attachments with raw bytes and X-Filename instead of passing base64 through model context.