---
title: What changed
description: See what each release of Turn Zero Cloud changed for you, newest first, with the library package versions it moved.
slug: /getting-started/what-changed
order: 12
visibility: public
redirects_from: [/getting-started/earlier-changes, /getting-started/changes-2026-10-23-to-2026-10-26, /getting-started/changes-from-2026-10-23-to-2026-10-26]
---

# What changed

Read what each release of Turn Zero Cloud changed for you, newest first.

## 2026-11-21 — build ad03b9b4b1

### Static sites

- #11349: A static site can now use `client/404.html` as its own not-found page. A missing GET or HEAD receives status 404, including paths that previously opened the home page. Leave this file out if a single-page application needs the root page to handle its routes after a reload.

- #11748 and #11750: Deployment progress and tool answers now say when no health or smoke check ran for a static site. Request the site's pages yourself; a completed deploy does not prove they work.

- #11751: Turn Zero's not-found responses now add `Cache-Control: no-store` when the response has no cache policy of its own.

### Account and deployment fixes

- #11755: Approving the deletion of your own account now clears that browser's sign-in and approval cookies.

- #11456: Text limits in tool calls now count Unicode code points, matching the stated bounds. Characters outside the basic Unicode range, such as many emoji, no longer use two places in that count.

- #11756: A redeploy during a database move now receives the settings needed for its health check, avoiding the earlier failed check.

### Guides

- #11776: [Let your app's users send feedback](/cloud/guides/let-your-apps-users-send-feedback/) shows a form and backend route that file users' reports in the application's issue space. Your AI tool reads the reports there; your users need no Turn Zero account.

- #11781: [Hand an app to another account](/cloud/guides/hand-an-app-to-another-account/) explains how to copy source, settings and exported data into a newly created application. It applies to finished work with no live end users; user accounts, sessions and passkeys are not exported.

- #9507: [Build a public upload gallery](/cloud/guides/build-a-public-upload-gallery/) now has its own page, with a complete server, browser page and check script. The [Storage guide](/cloud/guides/store-files/) keeps the signed-in upload example.

- #11639: The usage guide now names the dashboard's current Balance & usage headings, the group for applications no longer in the account, and each usage subtotal.

- #11675: Static-site cost guidance now links directly to the calculator's Static website example.

### Library packages and tzTrack

Production keeps build 137's library publication. This platform release publishes no newer library packages.

The tzTrack issue service's separate release serves version 0.86.0. This build does not publish that version to the library.

## 2026-11-21 — build 5513c6b672

### AI calls and usage

- #11703: An `allowance_default_unavailable` reply now says to wait about sixty seconds after the operator sets the default model, then retry.

- #11375: An AI call ended before the provider returns a status now draws nothing, except Gemini 4 calls, which retain an input estimate and no output tokens.

The [AI Allowance guide](/cloud/backend-packages/ai-allowance/) retains its earlier draw wording until its library update; the rule above applies to this build.

- #11491 and #11604: A large stored-file usage check can now resume where the previous check stopped, instead of starting again at the first file.

### Sign-in and email

- #11747: While signed out and sign-ups are open, the Sign up button stays visible beside the desktop Sign in menu and the phone’s navigation menu button.

- #10965: Passkey failures now give specific recovery steps, such as restarting a passkey step or reloading an outdated list.

- #11468: Automatic setup reminders and connection emails now leave room for your own Resend email request.

- #11508: A service notice no longer stops behind an account with many verified email addresses.

- #11480 and #11631: If the published welcome email is unavailable, its fallback now gives the current welcome guidance for the product you came from.

### Deployment and testing

- #10636, #11560 and #11745: Upload guidance now directs a tool without a shell to `read_status` with `brief: false`, then to the returned dashboard’s Upload a file control.

- #10843: Owner-only route guidance now compares a verified user address with `OWNER_EMAIL` taken from `read_account`. It explicitly rules out a shared admin key.

- #11395: On a local deploy from a signed-in connection with no authorized computer, `deploy_command` now returns a separate `authorize` step before the deploy command.

- #9950: Sign-in test advice now arrives with the recorded manifest when the app declares accounts and needs that check, instead of appearing for every newly created app.

- #11543, #11601, #11603 and #11615: Sign-in test guidance now states the audience and credential conditions together, including when a production test needs a minted token's grant.

- #11224, #11428 and #11501: Release guidance now checks changed application routes beyond the health path and reports writes left untested. It distinguishes test apps from apps with real users and prefers development for test writes.

- #10649: The version guide now explains how to compare `artifact_hash` with `turnzero hash`, including the first deploy's manifest change. It also explains when a serving version's `kept` file list is available.

- #11651: Size and plan refusals now explain that a serverless application's kind is fixed at creation. Work needing a server takes a new application, with its own address and data.

### Reading documentation and status

- #11135 and #11744: The status reference now lists the members kept in a brief answer in full. Waiting limits are described per serving process, rather than as one limit across the platform.

- #11367: An unknown documentation section now lists the nearest matching anchors first, so your AI tool can retry with one of them.

- #11400 and #11552: A documentation continuation now spells out the returned stamp alongside the next offset.

- #11402: The logs summary and diagnose guidance now point to existing router logs for complete server-error responses and to `read_status`'s `server_errors` for failed paths.

- #11616: Status guidance now tells your AI tool to read the operation's outcome, rather than treating a still-serving earlier version as proof that a redeploy succeeded.

- #11559: On an app with production alone, the manifest reply now puts `for: "local_run"` first in its `development_database` block.

- #11626: `read_plan_quotas` now lists only the measure values that its read accepts.

### Secrets and credentials

- #11426: The `list_secrets` summary now explains that an upstream key's null `applied_fingerprint` is normal: the gateway reads the stored value on each call.

- #11220, #11429, #11587 and #11588: Stored-key guidance now connects upstream declarations, settings bindings and restart requirements, and distinguishes a fixed host in code from one the app can change.

- #11653: The `mint_token` summary now makes clear that `session_credential_required` concerns the credential making the call, rather than the scope of the requested token.

- #11656: The log-retention reference now says the action refuses an application on the unlimited plan.

### Guides and cost estimates

A changed manifest's reply now includes `cost_reads`, pointing to `skill:cost-estimate` and `read_rate_card` before deployment.

- #11196 and #11538: [Build a database-backed service](/cloud/getting-started/build-a-database-backed-service/) now gives one path through building, checking development and releasing production, including the no-shell calls and sign-in guidance where needed.

- #11550: The command reference now links directly to installation instructions.

- #11680: The deployed-settings page now says that `TURNZERO_CLOUD_API` contains the scheme and host, with no path.

- #11685: The Database page now states that one migration file can contain several SQL statements and dollar-quoted function bodies.

### Library packages and tzTrack

Production keeps build 137's library publication, including `ai_allowance` 0.10.2 and `tztrack` 0.84.0.

No newer library copy is published by this release.

tzTrack’s issue service updates separately from the Cloud platform. Its own release now serves tzTrack 0.85.0; this platform build does not publish a newer tzTrack library package.

## 2026-11-21 — build 93210e2e7d

### Deployment and version history

- #11518: For a new server application, `create_application` now returns the next calls in order: submit the manifest, get the deployment command, and read the rates. The application id is filled in where needed; add your manifest before submitting it.

- #11472: When no compute exists yet, a console read during a deploy or promote now returns `deploy_in_flight` instead of `never_deployed`. The same applies during a first deploy upload. Follow the operation with `read_status` and `wait_seconds`, then read its outcome. If a new copy keeps stopping during its health check, the check can now end early and keep its startup error instead of waiting for the full deadline. A copy that is still starting keeps its normal check.

- #11197: Version-history rows now include `rolled_back_from`. A rollback names the production version it replaced; other rows return null. A rollback still has `kind: "promote"`.

### Clearer guidance for your AI tool

- #11515: The deployment summary now points to `deploy_command` for a local application folder when the AI tool has a shell and the command works there.

- #11528: The logs reference now connects a response's `x-request-id` header to the existing request-id filters. Console lines need a separate read and contain the id only if your code prints it.

- #11394 and #11571: The contract index now explains how to read a named file: call `read_context` with `source:` followed by its filename. A request using a bare contract filename also points to that readable id when it is available.

- #11549: Setup guidance now states together that a setup request starts the steps and that your own blanket approval covers step 2's changes, including adding the MCP server. The demo folder still needs approval of its full path.

- #11557: The external-API guide now names `TURNZERO_CLOUD_GATEWAY_URL`, with `TURNZERO_CLOUD_API` as the fallback, where code needs the gateway's address.

- #11562: The scheduled-job guide now opens with a four-step checklist for an app with production alone: declare, write the handler, deploy, and check both a manual and a scheduled run.

### Complete examples

- #11214 and #11403: The guide for bringing an existing app now includes one complete CommonJS Express server with startup migrations and text-ID records. It combines the connection retry loop, migration runner and record routes in one file.

- #11230, #11351, #11393 and #11553: The Storage page now includes a worked guestbook kept as one small JSON list. Its conditional writes retry a conflict instead of overwriting another request's message. The page also explains when to use a Database table instead.

- #11421 and #11516: [Build a mobile app's backend](/cloud/getting-started/build-a-mobile-app-backend/) now includes a fetch-based `/me` server without the Account package. It uses the same session-verification and verified work-account address rules as the package example.

### Library packages

The library serves `account` 0.23.3. Malformed session-token text returns a named refusal instead of throwing.

- #11578: Malformed verification-key text is now rejected consistently by the Account package.

## 2026-11-20 — build 62bb32deec

### Scheduled calls

Answers to scheduled HTTP calls no longer add data-transfer usage. The request or function call is still counted.

### Deployment guidance

With no authorized computer and a sign-in credential, `deploy_command` now opens its guidance with the `authorize` step.

The `deploy_files` summary now names it as the route wherever the command cannot run, or for a disposable simple test app's first deploy.

### Brief status answers

`read_status` with `brief: true` now includes `application.production_sleep`, so your AI tool can see whether production sleeps when idle.

### Command output

The `turnzero` command at version 0.53.4 replaces text-direction embedding, override and isolate controls in answered text with spaces.

Help now states separately that the command takes no credential in its arguments and which commands read no credential.

### Library packages

No library package version changed.

## 2026-11-20 — build 0373a54942

### Welcome page

When you arrive from a Turn Zero Tools or Image Gen page, the welcome page now puts Tools setup first. Arrivals from tzTrack and tzMesh keep Tools first; other arrivals show Cloud first.

If you press Resend email and the service declines your request with a reason, the welcome page now shows that reason.

### Markdown copies

An HTML page with a Markdown copy now names that copy in its HTTP `Link` header. Your AI tool can discover the Markdown address from the page's response.

### Library packages

No library package version changed.

## 2026-11-20 — build 99c1fdef0e

### Connect your AI tool

After you create an account, the welcome page offers setup prompts for Turn Zero Cloud and Turn Zero Tools. Copy a prompt into your AI coding tool to connect it. Connected products show a Connected mark and suggestions for what to try.

The welcome page also lets you resend the setup email. If your sign-in is too old, it asks you to sign in again first.

### Setup emails

If you have not connected your AI tool, we send the setup steps again about a day after the first email. We also send one email after you first connect. Replies to these emails now go to support.

### Your profile

Settings has a Profile page for your account picture when pictures are available. The Account menu shows the email address of the account you are signed in to.

### Default AI model

AI calls may now name `default` to use the platform's selected model and thinking level. If no default model is set or the gateway has not read one, the call returns `allowance_default_unavailable`. Name a `gemini-` model in the path, or retry after the setting can be read.

### Library packages

No library package version changed.

## 2026-11-19 — build cfa3b32c27

### Sign in

Choosing Sign in from a public page now returns an existing account to that page. A sign-in that creates an account opens your dashboard instead.

The signed-out header's Sign in menu also links to the Support agent and Documentation. It offers Create account when sign-ups are open.

Signing out now continues the remaining cleanup steps if one step fails. If it reports an error, repeat Sign out.

### Your first application

With no applications, your dashboard shows one card for connecting your AI tool and building your first app. Once a connection is found, the card shows it and explains how to start.

The sidebar's tzCloud menu opens tzMesh or the Turn Zero website.

### Deployment answers

`read_status` now says the deployment health check needs three HTTP 200 responses in a row. When server-error counts lag, the answer names those counts and gives a time to read them again.

If no computer is authorized, deployment guidance points to the command's `authorize` step before the deployment line. It also warns that a first deploy may set the application id in `manifest.json`.

Submitting a server manifest without a health path now returns `health_path_undeclared` while a deploy, promote or redeploy is in progress.

### Database answers

A `query_database` answer for a missing database now distinguishes a manifest with no database from one whose database has not been created. It names the first deploy or promote for production, or `submit_manifest` for development.

### Usage guidance

`read_rate_card` now explains how server awake time, a shared database and other usage contribute to an application's monthly total.

AI usage guidance now explains the fallback for a model the platform has not priced: the rates from the model with the highest base output rate.

### Library packages

No library package version changed.

## 2026-11-19 — build 34859ce195

### Set up tzTools

The setup pages for tzTrack and tzMesh now share the [Turn Zero Tools manual](https://turnzero.ai/tztools/). Use its [setup page](https://turnzero.ai/tztools/getting-started/set-up/). The old tzTrack setup addresses no longer answer after the site update.

### Pricing

The redesigned [pricing page](https://turnzero.ai/pricing) groups rates under "Hosting & Backend Rates" and retitles the estimate "Estimate your hosting cost". A link to tzTools' setup replaces the setup block.

### Usage and deployment guidance

`read_balance` now explains when usage reaches your balance. Awake time, stored data and transfer through the egress gateway or from storage areas post at the daily pass for the previous UTC day. AI calls, email, realm sign-in codes and the router's data transfer post as they happen.

`check_deploy` now states that its returned commands need Node.js 24 or later.

### Support email

If recording a support hand-off fails after an email was attempted, the answer keeps its send result instead of returning a server error. An unconfirmed send answers `not_sent` with `send_failed`; the email may still arrive.

### Library packages

No library package version changed.

## 2026-11-19 — build de73678434

### Sign in

After this update, sign in to the Turn Zero site once more in each browser, with your account, apps, settings and sign-in methods unchanged and your AI tools still connected.

Sign-in pages now omit the Apps menu. When you are signed out, their header also omits the Sign in link.

The Apps menu on invitation notices now follows each product's invitation requirement.

### Deployment guidance

Deployment guidance now directs your AI tool to the command when it has a shell and the whole app folder. Without a shell, it uses `deploy_files` within its limits.

### Publish static sites without a server

Publish web pages, styles and images as a [serverless application](/cloud/concepts/serverless-applications-and-servers/). No server of your own runs, so there is no size to choose and nothing to wake.

Your AI assistant calls `create_application` with `kind` set to `serverless` and no `size`. With a shell, `turnzero scaffold --template static-site` creates a starting site. Without a shell, your assistant can use `deploy_files` within that route's limits. A static site needs no root `package.json`. Follow [Publish a static site](/cloud/guides/publish-a-static-site/) for the steps.

A folder path serves its `index.html`. A page path with no file extension serves the root `index.html` when no file matches. This lets a single-page site keep its routes after a reload. Other missing paths return `not_found`.

Make the site public or admit only people you invite. Where custom domains are available, a production site takes one under the same rules as a server application.

There is no charge for a server, for serving a static-file request, or for storing the site's own files. Every file served counts as data transfer. See the [pricing page](https://turnzero.ai/pricing) for rates.

Serverless applications serve static files only. A zip containing `functions/` or `migrations/` is refused. A manifest declaring a database, a schedule or an egress host is also refused. Both refusals are `serverless_static_only`. Backend code takes a server application. Your AI assistant asks before choosing one.

### Library packages

No library package version changed.

## 2026-11-19 — build fd642e865b

### Sign in

Expired sign-ins show a page explaining how to start again. A sign-in started on the website lasts thirty minutes; an emailed code expires after ten minutes. A sign-in started from your AI tool and a link for adding another sign-in method each last ten minutes.

When you try another sign-in method for an existing account, the page names the method already linked and explains the next step. If undoing a passkey change fails, the page shows the failure's details when the service supplies them.

### Monthly charges

Each monthly charge shows the number of days its amount covers as its quantity.

### Library packages

The library serves these updated versions:

| Package | Version |
| --- | --- |
| `ai_allowance` | 0.10.1 |
| `email` | 0.4.2 |
| `logging` | 0.20.2 |
| `push` | 0.5.3 |
| `schedule` | 0.3.2 |
| `secrets` | 0.22.4 |
| `tztrack` | 0.84.0 |

## 2026-11-18 — build 3626b3055a

### Usage pricing replaces application plans

Your applications keep running. Each calendar month, your account gets Free Allowance and, for now, an extra introduction bonus. Usage is deducted from that balance at the rates on the [pricing page](https://turnzero.ai/pricing). You receive no bill, and prepaid credit is not yet sold.

### Your existing applications

Each application's production copy has a “sleep when idle” setting. It is on by default for Shared 0.5 GB and off for Shared 1 GB. Existing applications move as follows:

| Previous plan | New size | Production sleeps when idle |
| --- | --- | --- |
| Free | Shared 0.5 GB | On |
| Standard | Shared 0.5 GB | Off, as before |
| Pro | Shared 1 GB | Off |

A development copy always sleeps when idle. A server uses the balance only while awake; a sleeping server costs nothing.

### Logs, AI passes and a used-up balance

Logs are now kept for 7 days. Entries older than that are deleted at the switch. You can turn on the 30-day log option for an application.

tzTrack's AI passes stop in every issue space until its owner turns the passes option on for that space.

Emails and the dashboard warn you at 50%, 80% and 100% of the month's balance. At 100%, every application in your account pauses until the next month's first day. Your data is kept, and you can still sign in and export.

### Monthly items

A shared database, 30-day logs and tzTrack's AI passes are charged from your balance at the start of each calendar month. An item added during a month is charged for the days left. Removing an item or turning it off returns nothing to the balance.

An item your account already had is not charged in the month of the switch. Its first charge is on the next month's first day.

The shared database charge covers one application's production and development databases together. That charge stops only when you delete the application.

### Capabilities and limits

Every application can use email. Custom domains are not available yet. All sizes have the same limits except for WebSocket connections, whose limit depends on the size. Read the [usage guide](/cloud/guides/plan-and-usage/) for the limits.

A tzTrack account can hold up to 5 issue spaces and 100,000 open issues. Sharing a space needs no paid plan. tzMesh has no plan limits.

### AI use

An application has no monthly AI allowance. Calls through the platform's AI route deduct the provider's published price plus the pricing page's markup from your account's balance. Calls using the application's own stored key have no markup and deduct nothing from that balance. The first five chat images each month are free.

### AI tools and scripts

The tools change as follows:

- `set_app_size` changes an application's size and its sleep setting.
- `set_log_retention` turns the 30-day log option on or off for an application.
- `tztrack_set_ai_passes` turns AI passes on or off for an issue space.
- `read_rate_card` reads the rates.
- `read_balance` reads your account's balance and the month's lines.
- `create_application` takes a `size` and no longer takes `plan`.
- `read_status`, `list_applications`, `read_usage` and `export_account` answer an application's `size` and no longer answer `plan`.
- `read_status` also answers production's sleep setting, `production_sleep`.
- `set_plan` and `choose_tools_plan` are removed. The choice's refusal, `plan_choice_closed`, leaves with it.
- `read_plan_quotas` no longer answers `prices` or `subscriptions`.

Account reads and start answers drop `tools_plan`. Account reads also drop `app_hosting`. Account reads, space-member listings and sharing answers drop `tztrack_tier` and `tztrack_grace_end`. Issue-space listings drop `owner_grace_end`. The issue service's `tztrack_read_space` drops `tier`.

These refusal names are removed: `custom_domain_not_on_plan`, `email_not_on_plan`, `plan_custom_domain_conflict`, `free_application_limit`, `beta_plan_limit`, `plan_retired`, `socket_hours_capped` and `tztrack_paid_required`.

You need do nothing to keep your applications running. Turn on 30-day logs or tzTrack's AI passes if you want either option. If your usage would exhaust your balance, you receive warnings first.

## 2026-11-16 — build e2f3ab28c7

Your AI tool now follows clearer guidance on who may use a new application. Schedule reads explain an absent development environment, and status reporting covers production database servers.

Fixes that reached you:

- #9761. With nobody to ask and no audience stated, deployment guidance chooses public for code with its own sign-in, or invitation-only for code without it.
- #9968. Audience guidance now treats your project's stated intent, including its README, as your audience choice.
- #9821. A `read_schedules` request for development now explains when your application has no development environment.
- #10509. Status reporting now includes production database servers, so their outages can be reported alongside development database outages.

## 2026-11-15 — build 19602e9dbc

Setup and sign-in guidance is clearer. Activity reads now give your AI better timing information, and schedule reads help it work out how long to wait.

Fixes that reached you:

### Sign-in and setup

- #9702. Deployment answers and checks now warn when an invitation-only application has nobody able to sign in.
- #9976. When your verified address is known, the sign-in guide now shows an owner-only route that works from your first sign-in, without a placeholder or extra restart.
- #10334. Setup guidance now gives you a readable connection page when your AI cannot read the instructions in full.
- #10335. The connection guide now explains Claude's custom-connector availability.
- #10429. The mobile backend guide now links to a complete `/me` example using `fetch`, with no package to download.
- #10442. Sign-in guidance now explains how a valid application session lets same-origin browser requests pass the invitation-only audience check.

### Activity and scheduled work

- #10044. Usage reads now explain that a zero traffic total may mean requests are not counted yet, and tell your AI when to read again.
- #10445. A status read now tells your AI when to read server-error counts again if counting has not reached the window being checked.
- #10448. `read_schedules` now returns `read_at`; your AI can compare it with `next_due` to work out how long to wait.
- #9824 and #10528. Your AI is now told to get your agreement before halting production or running a production schedule by hand.

### Working with files and keys

- #10430. Key-rotation guidance now shows how to check the provider's response through the gateway. Success confirms acceptance without identifying the key value sent.
- #10435. Static-site guidance now says to keep your existing file names and relative links.
- #10450. The manifest's reply now distinguishes the platform's deployment check from outside health probes that need to pass the application's sign-in check.
