Self-Updating Client Trackers and Reports | hedwigai

• List workbooks — Every workbook this account has, with its title, the prompt of its latest run (`asked`), the folder it sits in, its version, when it last changed, and its state. `state` is one of: editing (a run is writing it now), needs_input (the document stopped on a question for a person), stale (a source it cites has changed since it was written), fresh (changed since you last looked), idle. Each workbook's id is what the other tools take as `workbook`.
• Search workbooks — The workbooks whose title, latest prompt or current text contain every word of `query` (any order, not case-sensitive), each with a snippet around the first match. Title matches come first, then prompt matches, then the most recently changed. A phrase such as 'tea importers' or 'weekly ops' finds the workbook it names.
• Start a new workbook — Makes a new workbook from EXACTLY ONE of: `instruction` — one sentence the way a person would say it ('find 30 US importers of loose leaf tea and the buyer at each'), which starts a run that researches and writes it, is billed to this account and takes minutes; `document` — markdown you have written, which becomes version 1 at once; or `sheets` — tabs, each {name, markdown}, which become version 1 at once. Writing content yourself runs nothing and costs nothing. Nothing that already exists is ch
• Edit a workbook directly — Writes a new version of a workbook yourself, with no run and no charge. Send `base_version` — the version you read — and EXACTLY ONE of `document` (the whole text of a one-tab workbook) or `sheets` (the tabs, in their new order: {id} alone keeps a tab unchanged, {id, markdown} and/or {id, name} replaces what you give, {name, markdown} with no id adds a tab, and a tab you leave out is removed). `note` is one line on what changed and labels the version. Every write is a new version by this account
• Put a workbook in your shell — Writes the workbook into your shell session's `/tmp` so you can work on it with bash. A workbook's tabs, and the spreadsheet they were derived from, live in this account's object store — the shell mounts providers, not that store, so this is the only way to open them from bash. It goes into THIS connection's shell — the one the `bash` tool runs in — unless `session` names another: a shell's id is what `session id` prints inside it. `/tmp` belongs to one session, so a checkout made in one shell
• Save what you changed in your shell — Reads back the checkout in your shell session and writes it as a new version. Send the `version` the checkout reported as `base_version`, and the same `session` if the checkout named one. A workbook made from a spreadsheet commits from `current.xlsx`: its tabs are derived from the file again, and the edited file becomes what the workbook is made of, so the formulas, formats and charts around your change survive. A markdown workbook commits from `tabs/*.md` — a file left alone keeps its tab, and
• Link a workbook to a file in Drive or Dropbox — Makes a file in this account's Google Drive or Dropbox the document of record for a workbook. The file is brought in now — an empty workbook is seeded from it, and one with versions gets its content as the next version — and from then on: - every run and every checkout_workbook brings the file in first, so work starts from what is in the drive, including an edit made in Excel or Word since; - every new version is written back over the file IN PLACE (the same Drive file, its sharing and revision
• Unlink a workbook from its file — Stops bringing a workbook's linked file in and writing versions back to it. Nothing is deleted: the workbook keeps every version, and the file stays in the drive as it is.
• Bring a linked file in now — Reads a linked workbook's file from the drive now rather than at the next run, and writes it in as a new version if its content changed — for after somebody has saved it in Excel or Word.
• Read a workbook — The current document as markdown, with its tabs (`tabs`: id, name, size — pass one tab's id as `sheet` to read just that tab), its version, its sources, the question it is asking if any, the list of versions behind it, and `messages`: the runs that were questions, each with its answer. This is the same document a person reads on app.hedwigai.com.
• Find a workbook from your own words — The one workbook your words mean, or none. Say it the way you would ask — 'Alterity's diligence memo', 'the weekly ops report' — and jev chooses among the account's workbooks by name, with 'none of these' always an option, so nothing is made up and no spelling has to match. `folder` limits where it looks: omit it for anywhere, `/` for workbooks at the top level only, or a folder path for that folder and everything in it. Answers `found` ({id, name, folder, and its title where it has one}) or nul
• Find a folder from your own words — The one folder your words mean, or none — jev's choice among the account's folders, with 'none of these' always an option. `parent` limits where it looks: omit it for any folder, `/` for top-level folders only, or a folder path for the folders inside it. Answers `found` ({id: the path, name, folder: its parent}) or null.
• Pick a table out of a workbook — One table from a workbook as data, chosen from your own words. Say what you want in `prompt` the way you would ask for it — 'the drug programs', 'who the competitors are' — and the table whose content that describes is chosen from every table in the workbook, by the headings above it as the document has them; you do not need to know a heading. The answer's `table` holds `columns`, `rows` (cell text in column order) and `citations` (the footnotes its cells cite, resolved), with `confidence`; `con
• Pick a section out of a workbook — One section of a workbook — the prose under a heading — chosen from your own words. Say what it is about in `prompt` — 'ATH434 in multiple system atrophy', 'the competitive landscape' — and the section that is about it is chosen from every heading in the workbook, as the document has them; you do not need to know a heading. The answer's `section` holds `heading` (the headings above it and its own), `level`, `line` and `text` (the section as written, subsections included), with `confidence`; `con
• List a workbook's sections — The sections of a workbook, each as `heading` (its title), `path` (the headings above it and its own), `value` (what is under it: `table`, `quote`, `list`, `block` — code, a chart or a diagram — or `section` for prose), `level`, `sheet` and `line`. Omit `prompt` for every heading. Give `prompt` — any words, 'everything about the Phase 2 programs', 'the risks and what mitigates them' — and jev weighs every heading against the others, by what each section is about, keeping the ones the words ask f
• Read sections by address — The sections an `address` from list_sections names, snipped out of the workbook's current version as written — a focused part of the workbook instead of all of it. Each of `sections` is as list_sections listed it plus `text`: its heading line and everything under it, subsections included. A section inside another one in the set comes once, inside it. `missing` names headings in the set that the current version no longer has. Reads only; no run starts.
• Find where an edit lands — Every section an edit changes, in every workbook it changes. Give the edit as `prompt` — any words, 'raise the Q3 rent to 2,400', 'the trial read out positive; update the program and the runway it buys' — and `targets`: the sections you already have, each a `workbook` with either an `address` from list_sections or one `section` (its `path` as listed, or a heading only one section has). It works for an edit or a bare topic ('finance'). jev weighs the sections against each other — every section of
• Ask a short question — A question whose answer is a word or a value, answered from this account's workbooks and programs without starting a run: 'Is Moderna NASDAQ:MRNA a biotech company with public float above $5B?', 'What is Alterity Therapeutics' lead program?'. `answer` is yes, no, unsure, invalid_premise (what the question names is not there: no such company, or a name and ticker that are two different companies) or the value asked for. Amounts are compared in code against the figure a program returned (`claims`
• Save a workbook's sources to the drive — Copies every web source the current version cites into the account's connected drive, in the workbook's subject folder — the same as Save to Dropbox in the app. Sources that are already files on a drive are left where they are. Answers with the drive and folder, how many were written and skipped, and any that failed; it stops at two minutes and says how many remain, and calling it again continues without copying anything twice. Refused while a run is editing the workbook. Nothing is billed.
• Answer what a workbook is asking — A workbook can be asking its reader something, shown as `asking` when it is read, one question at a time: whether to remove addresses a mail server refused, whether to remove citations to pages that could not be fetched, or whether to find sources for a document that cites none. `yes` records the decision and starts the run that does it; `no` records that the document stays as it is. Either way the answer is kept with the workbook, so nobody and nothing is asked the same question twice. `yes` st
• Ask a workbook to change — Starts a run on the workbook, described in one sentence the way a person would say it — 'add a section on Q3 headcount', 'refresh the cash figures from the latest filing'. The run is billed to this account and takes minutes; this returns its run id immediately and does not wait. The workbook shows what changed once the run finishes. An instruction that is a question about the workbook — 'which rows have no source?' — is answered instead, in seconds, from the workbook's own text: no version is wr
• Stop a workbook's run — Stops the run editing a workbook — the one read_workbook shows as `live_run`, or the `run_id` you name. The run stops at once. What it had written by then is not thrown away: if it changed the document, it becomes a new version, unfinished, and the version before it is kept and restorable; if it changed nothing, no version is written. A drawn procedure stopped before it finishes writes nothing. With no run going, this says so and changes nothing.
• See or change a workbook's settings — A workbook's settings, and changes to them: its `name`; its `title` — what it is shown by, where that is shorter than its name; its `mode` — `harnessroom`, `latticeroom` or `graphroom` (a harnessroom researches and writes; a latticeroom is handed a draft and marks each claim with its sources, and each reference with whether the work exists, leaving the words as written; a graphroom performs the procedure drawn for it in the app); for a harnessroom, the `model` its runs think with (one of the `no
• Bring back an earlier version — Makes an earlier version of a workbook its current one again, as a new version — the one it replaces is kept and can be restored in turn. read_workbook's `versions` lists them.
• Read a run's trace — Everything one run did, in order: each call to its model (`portal_returned`: the node, ms, prompt, completion and cached tokens), each tool it ran (`tool_returned`: the tool, its command, ms, whether it succeeded, the characters it returned), what it said, each draft it wrote (`drafted`: its bytes) and how it ended. Long text in an event — a draft's document, a line the harness printed — is cut to its first 400 characters and its length. list_runs gives the run ids.

Showing the first 25 tools — full list and schemas: Directory API.