Cobrain

Add text to the end of a note (creates it if missing). Prefer this over write_file whenever you are adding to a note instead of rewriting it: an append is atomic in the database and never conflicts, even with two writers at the same second. To change text that is already in the note, use patch_file. The brain is found from the path: omit workspace unless the tool asks for it.

Used by the Cobrain app: one note, rendered, with where it lives and who changed it last.

Used by the Cobrain app: notes whose text matches, with the matching lines.

Attach a real file (image, PDF, audio, anything) to the brain, next to a note: it lands in the attachments/ folder of the note's branch and a Markdown reference is appended to the note (link=false to skip, then place the returned Markdown yourself). Give the file in ONE of three ways: data (base64, files up to ~3 MB), url (a public URL, fetched server-side, up to 25 MB), or neither: you get a signed PUT URL to upload the bytes yourself (e.g. with curl), then call attach_file again with the returned name and confirm=true. Returns the brain path and the Markdown that references it.

Opens the Cobrain app in a tab beside the conversation.

Create a unit at path from the brain's template for type. type is free (studente, paziente, classe, cliente, progetto…): the template is _templates/<type>/ in the brain; if missing it is seeded the first time from the preset of that name (cliente, progetto) or, for any other type, from the generic template (_about.md with frontmatter status/context, _ai.md, decisions.md; placeholder {{name}}) — then shape _templates/<type>/ with the user so the next units of that type are born complete (see start_project). These names and keys are English in every brain, Italian ones included. On an existing folder it only adds the missing files (never overwrites) — use it to mark an imported folder as a unit. Then start_session and fill the sections following the for the AI comments in each section. In a Cobrain brain where you can write only a branch (shared with you by its owner) and not the root, the unit is still created from the default template.

Move a note, an attachment or a folder (with everything in it) to the brain's trash: the brain's owner can restore it from the Trash page of Cobrain for 30 days, then it is deleted for good. Previous versions come back with the note. Trashed notes still count toward the plan's note limit. Everything stays inside Cobrain. No tool can see or restore the trash.

The Dream: tidy up the brain's memory. Among the notes changed in the last N days (or all the notes of one folder) finds notes that say almost the same thing as another one, decision logs grown too long, [[links]] that lead nowhere, empty notes — and returns them with the step-by-step script: read, fix the safe things, tell the user what you found and ask once before anything that could lose content (merging different notes, deleting, superseding decisions). Use it when the user says 'dream', 'sogna', 'fai il sogno', 'tidy up', 'riordina', or pastes the sentence from Cobrain's Dream email or Dream page (that one names a saved pre-dream: pass saved=true).

Open a note by the id a search result carried, and get its whole text as JSON. If you are not ChatGPT, prefer read_file: same note, plus attachments and the version you need to rewrite it.

Notes carrying a tag.

Find notes, folders and attachments by NAME (part of the file name or path), like find on a filesystem. Use this to locate a unit's folder or a note you know by title, or to resolve a [[wikilink]] to a path.

List the notes (.md) and the attachments (with size) under a folder. Omit path for the whole brain. If the folder does not exist you get similar folder names.

Tags with counts.

The units in the brain (folders with _about.md: clients, students, patients, projects… any type): path, type, status (status: in _about.md: active, closed or free text), notes, last change, open sub-units. Closed ones only with all=true.

Your brain plus brains shared with you (id, name, access). The first line says which account you are connected as.

Everything under a folder (or with a tag) in one blob, INSTRUCTIONS FIRST: brain-wide rules (_ai.md at the root), the unit's _about.md and _ai.md, its context folders and the latest decisions are always included in full, then the notes of the branch (closed sub-units left out). Call this first when starting work on a unit. The header lists, by name, the notes that did NOT fit: open the ones you need with read_file instead of assuming the blob is the whole branch.

Rename or move a note or a whole folder (needs write access on both ends). Into another brain the attachments under it travel too. To move it into ANOTHER brain pass to_workspace (id, «mine», or the owner's name/email).

Opens the Cobrain app full screen from ChatGPT's sidebar: the user browses and searches their brain and attaches notes to the chat.

Change one part of a note without rewriting it: replaces old_text with new_text. old_text must be copied exactly from read_file (spaces and line breaks included) and appear once in the note; add a few surrounding words if it appears more than once, or pass replace_all: true to change every occurrence. An empty new_text deletes old_text. Only changes a note that exists. The brain is found from the path: omit workspace unless the tool asks for it.

Read a note. On an attachment path (see list_files) you get the file instead: small images inline, text files as text, anything else as a signed download URL valid one hour.

The most recently changed notes (who and when). Use it to catch up on what happened lately.

Find notes by words in their text, across your brain and the brains shared with you. Returns note paths with the matching LINES as excerpts (marked with …). A hit from a shared brain is tagged [<owner>'s brain · workspace=<id>]: pass that workspace to read_file/load_context/start_session for it. Excerpts are not the note: before answering from one, call read_file on its path; for the surrounding context, load_context on the note's folder. Each hit carries the note's date (from its frontmatter or file name, else when it was written). Hits under a Maybe related line match only SOME of the words: prefer the ones above it. Use 2-4 keywords, not a sentence. path narrows the search to one branch (a client, a project): use it whenever you already know where you are working.

Used by ChatGPT's composer to let the user @-mention a note (typeahead by name). Other clients: use find_notes.

Used by ChatGPT's plugin settings page: the settings, their current values and layout.

Used by ChatGPT's plugin settings page to save the settings the user changed.

Guided brainstorming to set up a new unit in the brain — a client, a student, a patient, a project, whatever the user wants to track: returns the step-by-step script to follow with the user (what to track and what to call it, a few questions, structure proposal, create_unit, filling). Use it when the user says 'nuovo cliente', 'nuovo progetto', 'nuovo studente', 'start project' or wants to add someone/something to the brain.

The one-minute briefing on a unit (a folder with _about.md: a client, a student, a patient, a project… whatever the user tracks): brain-wide rules, its _about.md, _ai.md, context folders, latest decisions — in full — then its open sub-units and the last 15 changes. Call this FIRST when the user names a unit; then load_context on the sub-folder you need. unit is a path or a name. Special file names and frontmatter keys are English in every brain: facts about the unit go in _about.md (frontmatter status: active|closed, context: folders always loaded), decisions in decisions.md (append_file), templates in _templates/.

Create a note, or REPLACE its whole content (needs write access). Adding a line to a note that already exists? Use append_file. Changing one part of it? Use patch_file — write_file with a short text erases everything else. Replacing an existing note: pass base = the version read_file printed at the end of that note, so you never overwrite what someone else (a person, another AI) changed since you read it — if it changed, you are told to re-read and merge. The brain is found from the path: omit workspace and pass it only when the tool asks for it (a new branch, or one present in several brains), with the brain the user chose.