Record messages
record_messages · Changes things
Keeps messages between this person and you in their Trillion account, word for word, so every other AI they use can read them (read_messages). Call it on every turn, before you send your reply, with both sides: the person’s message (speaker “person”) and the reply you are about to send (speaker “ai”), then send exactly that reply. Each message: conversation_id (the same id for every turn of this chat; make one up at the start if the app gives you none), app (your app, like “ChatGPT”, “Claude”, “Muse”; it defaults to the app you connected as), speaker, text (exactly as written, never summarized), at (leave it out for this turn and Trillion stamps the time; give it only for older messages you are sending from an earlier conversation, as their real ISO time), model (optional) and about (optional: what it concerns, as [{kind, id}], like {“kind”: “pursuit”, “id”: “pursuit:content”}, {“kind”: “decision”, “id”: “<case_id>”}, {“kind”: “tool”, “id”: “tool:extract-sound”}, {“kind”: “file”, “id”: “1f:…”}). Up to 100 messages a call, so an earlier conversation can be sent in batches. A retry with the same original at changes nothing; keep the original time when retrying. Without at, the time is when this call records it. Returns how many were new and each message’s id. It does not send the messages to anyone.
Inputs
Example
Read messages from all AIs
read_messages · Reads only
Reads messages this person exchanged with any of their AIs, newest first: each with id, conversation_id, app, speaker, text, at and about. Filter by conversation_id, by app, by not_app (everything except one app, like your own, to see what they said elsewhere), or by about_id (messages about one pursuit, decision, tool or file). Returns a page and next_before; pass next_before as before for the page before it. Read it at the start of a conversation to pick up what the person has been working on with their other AIs. It never changes anything.
Inputs
Example
Bring in one chat
bring_in_chat · Changes things
Brings one earlier chat into Trillion from its exported archive, word for word, with the files its messages carried: into the person’s private Trillion (leave space_id out) or into one shared space they name (space_id). Files are found among the person’s own files and never copied: attachments maps each attachment id in the chat (like file-AbC123) to the name or identity of that file in their files, and attachments_folder names a folder whose files start with those ids, as in an export. In a space, other members can open those files while the person stays in the space. Files that could not be found are listed in missing_attachments; put them in the person’s files and bring the chat in again to add them. Bringing the same chat in again adds nothing twice. For a whole account export use bring_in_chat_export.
Inputs
Example
Bring in a chat export
bring_in_chat_export · Changes things
Brings a whole account export of chats (for ChatGPT, conversations.json) into Trillion, from a file already in the person’s files (file: its name or identity), with the files the chats carried: into their private Trillion, or into the one shared space they name (space_id). Upload the whole export folder, not just conversations.json: its files are found in the same folder as conversations.json and the folders under it, or in attachments_folder if the person put them elsewhere, and are never copied. In a space, other members can open those files while the person stays in the space. Files that could not be found are listed in missing_attachments; add them to the person’s files and bring the export in again to add them. A page of conversations per call; call again with offset set to next_offset until it is null. Nothing is brought in twice. Needs the private reach, since the file is private.
Inputs
Example
Read what my AIs did
read_activity · Reads only
What the AIs connected to this account did, newest first: each row says which AI app did it, what it did (handed off work, ran a tool, saved a file, published a tool…), the item it made or touched, whether it worked and, if not, why. Use it to see what another AI already did before repeating it. app narrows to one AI app by the name shown in its rows; for narrows to what was done for one thing (a decision’s case_id, or an id a message was about); before (a row’s seq) reads older rows; limit caps how many (default 50). Each row lists the ids it used and what it was for; trace_item follows them. It changes nothing.
Inputs
Example
Trace how something came about
trace_item · Reads only
Why did this happen: the chain of AI calls, tools and files that produced this item, and what used it. item is any id a row of read_activity shows (a file’s identity, a handoff_id, a case_id, a task id, a tool id). Returns steps in order: the calls that made it, then the calls that made what those used, hop by hop, then the calls that used it or were done for it. Each step is a read_activity row plus link (made, used or for), about (the id that step is for), because (the seq of the step that led to it) and depth. It changes nothing.
Inputs
Example