0.13.0 — see and test push notifications

Push notifications reached the API, so the tools can see and test them.

A live connection reaches a phone only while the app is on screen; the operating system closes it the moment the app goes to the background. What actually reaches someone whose app is closed is a push, and it now exists as a send_push automation action.

Added

• push_status. Whether an app can notify phones, whose Firebase project it sends through, and how many devices have registered. When nothing is set up it says how to set it up — and says plainly that the service account is a private key which must not be pasted into a chat.
• send_test_push. One real notification to named app users, so the chain can be proven before an automation depends on it. The answer separates "delivered", "this user has no device yet" and "the service says this device is gone", because those are three different problems.

On the API

POST /apps/{slug}/api/devices registers a phone; POST /organizations/{org}/apps/{id}/push/credentials takes the customer's own Firebase service account, encrypted at rest; send_push is a new automation action that defaults to the row's owner.

0.12.0 — query_items can filter, sort and search

query_items could only page. Anyone who wanted "the workouts assigned to me that are not done, newest first" had to fetch every row and filter on their side — slow, expensive, and it reads like the product cannot query its own data. The REST endpoint behind it always accepted search, status, priority and a sort field; the tool simply never passed them.

Added

• query_items takes search, status, priority, sort and archived. search matches the row title and its text cells; status and priority take a comma-separated list; sort takes title, createdAt, updatedAt or status with a leading - for descending; archived chooses active (the default), archived or all. They apply to the paging loop too, so an unpaged call comes back narrowed instead of pulling every row first.

Changed

• query_items now says in its description that a published app's own REST endpoints take a fuller grammar — nine filter operators per column, relation filters, per-field search — and points at get_app_spec. The tool is the admin view of a board, not the app's query language.

Alongside this, on the API

Sorting by updatedAt is new on the server too. The App API also gained the three things a client needs to work offline: ?since= (what changed, plus the ids that were deleted, plus a cursor), Idempotency-Key on writes, and POST /<endpoint>/batch for up to 50 operations. All three are in each app's generated OpenAPI.

v0.11.0 — delete_project, records not tasks, honest errors

From a reviewer's report on API quality.

Added

• delete_project. Boards, columns and rows could be deleted and projects could not, so anything built for a test stayed forever. The project goes to the organization recycle bin and can be restored from the admin.

Changed

• query_items on a data board returns records, not tasks: status, priority, dueDate, subtaskProgress and the rest of the task fields are left out, matching what the App API already does for those boards.
• build_backend says when a row has no title and no text value, so nothing can reference it, instead of reporting the missing title as a problem in the board that pointed at it. The description now matches the code.

v0.10.1 — build_backend cleans up after itself

Fixed

• build_backend no longer leaves a half-built project behind. When the plan does not allow another published app it says so before creating anything, and if any later step fails it removes the project and the app it made instead of naming them in an error message.

Server-side fixes shipped the same day and need no client update: links made when a row is created are now written to the relation graph, a deleted row never comes back as a live link, and a required column cannot be emptied through any channel.

v0.10.0 — deploy a frontend from ChatGPT

Added

• deploy_frontend takes the site in one of three ways instead of only a local folder: files (inline path + content, base64 for binaries; up to 500 files / 8MB) so ChatGPT and every hosted client can write a page and put it live in the same turn; zipUrl (a public https zip such as a Lovable/Bolt export or a GitHub release asset, up to 50MB; a single top-level folder is re-rooted); and dir as before. Exactly one is required. Private hosts, plain http and non-zip answers are refused before anything is uploaded.

v0.9.0 — data boards

Requires a TaskLite server from 2026-09-06 for the new behaviour; older servers ignore kind.

Added

• Data boards. create_board takes kind: "tasks" (default) also gives the board the built-in task columns; "data" creates a plain table with only the columns you add. build_backend boards default to "data": a backend's tables are customers, orders and payments, not to-dos. Rows of a data board come back from the App API without status and priority.
• Relation values from the App API always carry relatedItems: [{ id, title }] next to relatedItemIds, so a frontend shows a name without a second request.

v0.8.3

build_backend: sample rows are created in dependency order, so board order in the spec no longer matters; a row linking to a title absent from the spec is refused before anything is built. See CHANGELOG.md.

v0.8.2

build_backend: relatedBoard and relationType are typed fields on relation columns (relationType is an enum with business-model wording, echoed in the summary). See CHANGELOG.md.

v0.8.1

build_backend links boards: a relation column with settings.relatedBoardName is wired to the named board, and sample rows can fill relation cells by the related row's title. The API summary echoes scopes. See CHANGELOG.md.

v0.8.0

build_backend — one call builds a project, its boards, typed columns, optional sample rows, and optionally a published REST API with an endpoint per board and a server-side key. The model designs the schema; the tool executes it and returns one compact summary.

Organization defaulting for OAuth users (ChatGPT, Claude web): the single writable organization is picked automatically; with several, the error lists them so one turn suffices.

create_app_api_key accepts scopes; the reserved alias names are spelled out on the endpoint tools.

Tools 44 → 45 (hosted 40). See CHANGELOG.md.