What's new in Product Printer

A project's chat is now one shared conversation for the whole team, the Product agent and the AI CTO.

Write @AI CTO to ask the AI CTO a technical question; otherwise the Product agent answers. When an agent proposes something that changes the project or costs credit, it waits for your Validate. Conversations you had alone before stay private and no longer show in the project chat.

In a project's chat, each answer now shows who wrote it: the Product agent, with its name and avatar.

Hover the name to see what it handles. This prepares the AI CTO, which will join the same chat.

Traceability now says "Proven" only when what you asked for works on the app that is running.

Each acceptance criterion shows its automatic test result and its live check on the running app. A requirement reads "Proven live" only when every criterion passed a live check on the release that is running; a passing test alone reads "CI only". The page header shows "Proven live: X/Y requirements" for that release.

If nobody can sign in to an app hosted by Product Printer, an admin can now give its first admin access from the project's Deployment page.

Product Printer adds the project contact (and, if you tick it, this instance's admins) to the app's list of allowed users and makes them admin, then shows what it changed here and in the project's activity. Nobody needs to touch the app's database.

Builds now always let your app's first admin in, and keep a step's acceptance tests unchanged through the agent's own review rounds.

A sign-in list or "authorized users" list that a build adds now starts with the app's first admin, so nobody needs database access to get in.

A requirement that repeats another can be marked "Duplicate of FR-x", even before your first release, and the mark can be undone.

A duplicate is not planned on its own, coverage shows it merged into the original, and "Update the plan" lists likely duplicates instead of adding worksteps for them. Owners, maintainers and admins mark it from the requirement panel or over MCP (mark_duplicate).

Over MCP and in the chat, a plan update can now add only the new requirements you pick, in your order, each after the worksteps you name.

Pass `add` to `propose_plan_update` and `apply_plan_update`, e.g. FR-57 after WS-7, then FR-56 after FR-57. New requirements you leave out stay unplanned and are listed as such.

When AI stops because a budget is used up, the message now says whose budget it is, how much of yours is used, and who can raise it.

If Product Printer's shared trial budget is the cause, the message says so, says your own AI budget is not used up (for example "€50.71 of €85.69 used"), and says only Product Printer can lift it. If your own AI budget is used up, the message shows the amount and says Product Printer sets it: an admin can connect their own AI, or ask Product Printer to raise it.

The production line now stops after a build runs out of AI credit, instead of starting the next workstep.

Before, the next workstep would fail the same way and use up one of its attempts. The line starts again once someone runs a build again, for example a retry of the stopped workstep.

In a project, you can now fold the chat on the right away to give the page more room, and open it again in one click.

Use the fold button at the top of the chat, or Ctrl + . (⌘ . on a Mac). Your choice is remembered in this browser.

The "How likely are you to keep using …?" question no longer covers your work: it is a small line in the side menu, and the form opens only when you click it.

It appears for one session, then not again for 7 days ("Not now" does the same).

The project menu is now grouped along the way you work: Define, Build, The app, People & access, Project settings.

Worksteps and Traceability are tabs of the Production line. Under People & access, Members are the people working on the project, App users are the accounts of the app you build, and App credentials are your client's API keys and tools. Assigned Team is gone: it showed the same people as Members.

"Replan everything" now shows the estimated cost of the builds even when the planning cost is not recorded.

It says that the planning run is not included, instead of showing no estimate at all.

A requirement change now updates only the worksteps it touches, after you review its impact.

On the Worksteps page, "Update the plan" shows what changed, which worksteps it affects, the tests and design decisions involved, and the estimated AI cost. Worksteps not started yet are updated; built ones are never rewritten, a small follow-up step comes after them. "Replan everything" stays available, with a warning and its cost.

Change requests and requirement changes now wait for your app's first release, so work already built is not redone.

The lock starts when the first build starts and lifts as soon as the first release is deployed. Before the first build, you can change requirements freely, and support requests are always open. The change request form, the Requirements page, MCP tools and the assistant all say why and when it opens.

Your app's setup now runs to the end by itself: no step-by-step "advance" is needed.

When it truly needs you (for example the e-mail of the app's first admin), it says exactly what and how to fix it, in the app and through MCP. A new project via MCP asks for its name and contact e-mail up front.

The assistant now answers "how do I link a GitHub repository?" correctly: during the trial the app setup creates the repository for you.

New projects ask for the contact e-mail (the app's first admin) right away, and say who owns the code repository during the trial. The assistant also no longer looks up a wrong project.

On a hosted trial, the Deployment page no longer shows a Supabase dashboard link that only opened an access error.

It says instead that Product Printer hosts the app's database during the trial, and how to get a direct database change made.

Testing a Supabase key under a project's Connections now works with an anon key too.

You can now add your client's own API keys (OpenAI, Supabase, any API) and MCP servers to a project, so its assistant can use them.

Find it under the project's Connections tab. Keys are stored encrypted and only their last 4 characters are ever shown.

"Propose a design" now works on large projects: the design is written in parts instead of failing at the AI model's output limit.

Each part shows on the progress card ("Part 3 of 7: data for Billing"). Small projects still get their design in one go. If the AI model still cannot finish, the message says what to try next.

Committing worksteps no longer fails because the AI miscopied a requirement reference.

If a proposal still points to a requirement the project does not have, the page names the workstep and offers to generate the worksteps again.

New instances come with a demo project, Mini CRM for a fictional client, that walks you from an intake interview to a built app and your own change request.

On an existing instance, add it from the admin overview with "Add the demo project".

Generating worksteps now finishes even when the PP Orchestrator needs several minutes, instead of stopping with "The run stopped responding".

While the run waits for a free PP Orchestrator slot, its progress says so. If it still cannot finish, the message says why, and the Retry button stays on the Worksteps page after a reload.

Syncing CI secrets now works for project repositories only and no longer needs a personal access token.Self-hosting

The admin change request list now reads GitHub with your instance's GitHub App, so it no longer needs a personal access token.Self-hosting

Each organisation in `CR_SEARCH_ORGS` (default: the organisation connected in Settings → GitHub) is searched with a read-only token of the App. An organisation where the App is not installed is skipped and named in the function log. `GITHUB_TOKEN` is no longer read by this list.

New projects are always set up with your instance's GitHub App; the older setup through a platform repository and a personal access token is removed.Self-hosting

The French version of the app is rewritten in plain French, with the same words as the website: besoin, tâche, demande d’évolution, mise à jour, en service.

Developer terms (pull request, merge, build, CI) are replaced by plain labels such as « réalisation », « intégrée » and « contrôles automatiques ». French e-mails and both training courses follow the same rules.

Labels that always showed in English now follow your language, and the sign-up page and sign-in e-mails exist in French.

The recent What's new notes are in French too; a note can now carry its own French text.

In French, requirement references now read B-01 (besoin), Q-01 (qualité) and CV-01 (condition de validation) instead of FR-01, NFR-01 and AC-01.

The stored references do not change: links, searches and the support form accept both.

Instance deploys now retry a brief database connection failure during migrations instead of failing the whole deploy.Self-hosting

The migration step makes up to 3 attempts when Supabase's connection pooler times out, waiting 10 and then 20 seconds. A failure now shows the database's own error message.

The manual rework editor image now ships the fixed undici (CVE-2026-109869) in VS Code, npm and pnpm.Self-hosting

Trainings now show their version and what changed, with "New" and "Updated" marks on lessons that changed since your last visit.

Each course page has a "What's new" panel. A lesson you completed before it changed keeps its done mark and says "Updated since you completed it", with a link to the changes.

Training: the course “AI for SMEs” is now version 1.1. See What's new on the course page.

The trainings now explain the difference between a single AI step and an agent, with a short quiz to tell automation, AI steps and agents apart.

In AI for SMEs, module 05 "Agents and guardrails" covers what an agent may and may not do and how it asks for new tools through a change request. Admins get two new lessons: "A single AI step or an agent" (Selling AI to SMEs) and "Scoping an agent: its permissions are requirements" (platform training). The agent is described as an optional add-on, offered as a pilot.

Generating worksteps no longer stops at the model's output limit on larger projects.

The plan is drafted first, then each workstep's instructions are written on their own, and progress shows how many are done.

When Product Printer's monthly trial AI budget is used up, the app now says so plainly instead of asking you to contact your administrator.

On a free trial you are the administrator. The message now reads "Product Printer's trial AI budget for this month is used up — we've been told and will sort it out." AI comes back by itself once the budget has room again.

The training brief is now titled in its own language, listed under Documents, and "Refine with agent" starts from it.

Sample projects of a failed setup check ("PP setup check") are now removed even when nobody checks on that run again.Self-hosting

Record a client's existing tools and app during discovery: Requirements → Refinement → Existing tools & apps.

Tick the tools they use and answer two questions per tool (embed or data, one account or each user's own). If they already have an app, copy the prompt for their tool and paste back its summary: the next requirement extraction uses it. Keys and passwords are refused; only their names are kept.

The app and its training courses now come in 19 languages: English, French and German, plus 16 machine-translated languages such as Spanish, Polish, Arabic and Japanese.

Choose yours under Account → Language; machine-translated languages are marked there and on the course pages, and the English text is the reference. The admin courses are now also in French: until each lesson has been reviewed, it carries the same machine-translation note. Legal text stays in English, French and German.

For every tool a client names in discovery, the requirements now ask whether to embed it or use its data, and whose account it uses.

When the project records the client's existing tools or app, the draft requirements can include a data migration from the previous app and the tool connections, marked as coming from what the client already has.

Worksteps no longer send the coding agent to files your app does not have: plans are checked against the app's real files, and a file a step creates must be marked as new.

A proposed breakdown that names a missing file shows a blocking finding you can resolve in the chat; a change request's workstep gets a note telling the agent where to look instead.

Every app's pull requests now prove its database changes apply on an empty database, so a broken migration stops before it is merged instead of failing the release.

Existing apps get the check with their next template update.

On projects hosted by Product Printer, the Playground can check and merge a change again, and it carries your instance's name instead of "AvoDev".

When the repository access is missing or expired, the Playground now says so plainly instead of blaming the GitHub App. The new name shows once the project's playground is set up again or updated.

A workstep shown as done while its pull request was never merged now goes back to "needs attention" with the reason, instead of sitting there as shipped.

This happens by itself when you open the project's Overview, Worksteps or Production line; the line then stops calling it merged, and you can check the gate again or retry the build from the workstep.

Each acceptance criterion now shows its test-first proof: its test failed before the feature was built and passes with it, with links to both test runs.

A criterion that a person has to check, such as a look or a wording, shows "Human check pending" on Traceability until the project owner or the agency signs it off. Until then the requirement is not shown as verified.

While your app is being built, its home page shows "Coming soon" to visitors, and signed-in users see the build steps and where each one stands.

The page goes away by itself once every step is live. Apps created from now on have it.

A new project page, "What's in the app", shows the parts every app comes with, what has been built for you (with its pull request) and what is still to come.

Find it in the project menu, next to the production line.

A project now shows "Building — 3 of 6 worksteps shipped" until everything in scope is released, and "Live" only after that.

You see it on the project overview, in your project list and on the admin client board.

New apps show their home page again instead of a blank screen, and no longer log errors on load.

Apps created earlier keep the old home page code; their CI now runs a "Pages Smoke" check that flags it.

Releasing a version of an app hosted by Product Printer now starts its deploy to production; before, the release was published but its deploy never started.

A release of an app hosted by Product Printer now goes live: its database changes are applied first, then the site switches to it.

Releases shows where each deploy stands (requested, deploying, live, or not deployed with the reason), and the production line shows which version is live. Admins can deploy an earlier version again; the site goes back, the database does not.

Smaller fixes from the second canary walk: the requirements list is sorted by key, narrow layouts stack instead of squeezing, and traceability shows "–" until something is built.

The "How is it going?" question now comes after your project's first build or on your second day, not in the first minutes. The empty "no conversation yet" text no longer shows above a captured discussion.

Fewer clicks to a plan: "Accept all proposed" accepts every proposed requirement at once, and your answers to open questions now go into the requirements by themselves.

Design decisions show when your answer is saved. A proposal that needs a closer look still waits for your Merge.

The activity log on a change request now reads as sentences, such as "The change request was approved.", in your language.

Worksteps now show their plain-language title and summary, and their requirement chips use the same keys as the Requirements page (FR-01, NFR-01).

The technical title stays under the heading of each workstep.

The solution design and the proposed worksteps now open with a plain-language summary you can read with your client.

The technical details for the builder are still there, folded under "Technical details". Underlined terms explain themselves when you tap them.

Asking the assistant which change requests a project has now lists all of them, including approved ones.

It only limits the answer to the requests waiting for a decision when you ask for those.

"Refine with agent" now keeps working if you close the panel or leave the page: the draft requirements and open questions appear in the list when it is done.

Your notes are saved as a discussion on the Requirements page, like a recorded or pasted one.

Worksteps now build on their own: once every requirement in a workstep is accepted, it goes to the PP Orchestrator without a click, one build at a time per project.

This applies to requirements accepted from now on by an admin, owner or maintainer. Earlier work, backlog worksteps and any build that failed or was cancelled keep the Execute button. A change request still waits for the client to accept its price.

New Production line page in every project: watch each requirement move from defined to accepted, queued, building, in review and verified, live, with the build log beside it.

A requirement counts as verified only when its tests prove every acceptance criterion. Everyone on the project, the client included, can see the page.

A deploy no longer fails at "Auth emails (templates)" when Supabase briefly returns the old auth config right after updating it.Self-hosting

The step now reads the config again for a few seconds first. When a setting really does not stick, the error lists each one with what was sent and what was read back.

A new app's database setup no longer stops at the demo tenant step on current Supabase versions.Self-hosting

The demo accounts are now created in a way that matches how current Supabase Auth keeps email addresses unique. CI now builds a new app's database from scratch on every template change, with an app migration that uses the tenant helpers, so a template migration that cannot apply is caught before release.

At the end of the AI for SMEs course you can turn your process into a new project, or add it to an existing project as a change request.

Find it in your project under Training notes. A new project starts as a draft with your notes as its brief. A change request carries your notes and the requirements you propose; the team prices it and you accept it before anything is built.

Build steps now follow the security checks' data-access rules from the start, so their pull requests pass the security gate without a second round.

When a step's acceptance criteria can't be mapped to tests, the step now says why instead of leaving the mapping empty.

Your account, My projects, My feedback, New project and the training course now open inside the same menu and layout as every other page.

The Feedback box now has the same dictation microphone as the chat, wherever speech-to-text is set up for your workspace.

A new instance now looks like Product Printer: black and white, with colour only where something needs your attention, in light or dark mode as your device is set.

Your own colours, fonts and logo from Settings → Branding still replace it.

Every page inside a project now has the same layout as the rest of the app: lists use the full width, forms one readable width, always left-aligned.

The change request pages no longer draw their own top bar, and the rework bench uses the full height of the page.

New project agents run a tested, pinned OpenClaw release instead of whatever "latest" is that day.Self-hosting

The default is ghcr.io/openclaw/openclaw:2026.10.5 pinned by digest. To run another release, set the edge secret OPENCLAW_DOCKER_IMAGE (tag@digest), and run the setup check before relying on it.

After paying or managing a subscription in Stripe you now land back on the project's Billing page instead of a "page not found".

On your first day and after you set up your first project, Product Printer asks "How is it going?" so you can tell the team in one line what is broken or missing.

Your answer is sent as feedback to Product Printer: you can follow it under My feedback, like anything you send from the menu. "Not now" puts the question away; the menu entry "Feedback to Product Printer" stays available at all times.

Training now has its own place in the sidebar, next to Projects: the AI for SMEs course, with your progress kept per person.

Admins find three self-service courses under Admin → Training (AI for SMEs, how the platform works, Selling AI to SMEs). What a team writes about its own process during the course stays in the project, under Training notes; old project training links open the course.

You can now send Product Printer a problem, a change you need or an idea about this app, and follow their answer under My feedback.

Open it from your account menu ("Feedback to Product Printer"), or ask the assistant in the chat: it fills the form and you confirm before anything is sent. You can attach a screenshot and, if you agree, a few technical details (page, version, browser, recent errors) — never client data.

Plainer words for advisors: the Requirements page, the build pipeline and error messages now say what to do, and "AI isn't switched on yet" comes with a button to switch it on.

Admin → PP Orchestrators now leads with "Your PP Orchestrator — hosted by Product Printer (included)"; running it on your own computer or server is under "More options". Approving a change request before the app is set up now links to the Admin dashboard.

On a phone, a project page now has one menu and room for the project name; the Account page shows Telegram and Billing only when your workspace has them.

The project menu has a "Workspace menu" entry for everything else. On a change request, the decision form gets the full width, and the price field shows an example ("e.g. 1250") instead of a price.

Setting up two-factor sign-in works again after a cancelled or reloaded attempt, and the invitation page picks up where you left off.

"You're now logged in" now appears only after the 6-digit code, and after a reload the invitation page goes straight to the authenticator step instead of asking for the password again.

Your language is now a setting of your account: choose it once in Account settings and it follows you on every device you sign in on.

The language switcher is gone from the menus once you are signed in. E-mails we send you use the same language where they are available in it, and notifications now also reach you on Telegram when you have linked it.

Every pull request of a client app now gets an automatic design check: overlapping buttons, sideways scroll and broken images block the merge.

The check opens the app at phone, tablet and desktop widths and comments a table with screenshots on the pull request; the playground shows "Design check: N issues" with a link. Clipped text, small tap targets, contrast, missing alt text and console errors are warnings. Known exceptions go in `.avodev/design-check.json`, each with a reason.

A change pushed straight to an app's main branch, without the merge gate, is no longer deployed: the project shows "Unreviewed change on main" with Revert and Review.

Revert opens a pull request that undoes it; Review creates a change request, and approving it lets deploys resume. Existing apps get this with their next template update.

When your agency prices a change request, you accept the price on the request before it is built, and each promised result is checked by a test.

The workstep waits until you click Accept price; declining cancels the request. Projects without a client of record keep booking the price at approval.

Who is responsible now decides who merges and who pays: clients can ask for help or a review, and each project shows its AI usage and request costs under Billing.

Agencies set the AI margin, support analysis fee, warranty and "Client deploys themselves" under Admin → Billing (per project in Settings). In the playground, "Ask agency to review" makes a priced change request; "Get help" opens a support request: an analysis fee you confirm, then a fix quote. Existing projects: a Champion merges only once "Client deploys themselves" is on. Clients see billed amounts only; AI cost and margin stay with the agency.

AI usage on a subscription or an unpriced model is now billed at its list-price equivalent, the client's own AI is never billed, and a review's price is booked only when the client accepts it.

Admins keep the list prices under Admin → Billing (each change is a new version) and mark a client's own AI under Connect your AI, on a PP Orchestrator that serves only that project. Clients accept a price with one click on the change request.

The playground now turns each wish into a requirement you confirm, then builds it with one test per acceptance criterion, ships it as a pull request and merges it when CI proves it.

Merging deploys, so only an admin, a project owner or maintainer, or the project's Champion can merge. The Merge button stays off until CI is green on the latest commit, every acceptance criterion has a passing test, and every commit names its requirement. A change to an existing requirement goes to your agency as a change request first, and you see their price before it can be merged.

Refining an epic step and the legacy GitHub orchestrator deploy now answer a bad or signed-out request with a clear error instead of crashing.Self-hosting

Before, both failed with "Function exited due to an error" on every call when an optional setting was missing.

Change requests now show their pull request, say in plain words when a build is on hold, and send status e-mails.

The agency's admins get an e-mail for each new request; the person who filed it hears when it is approved, in progress, on hold, merged and live. E-mails go out through the instance's own sender.

Every new project now belongs to an organisation: pick one, or create it on the spot, when you create the project.

Admins can open an organisation to see its projects and start a new one there, mark one organisation as "our company" (the default for new projects), and filter or group the project list by organisation — projects without one get an "Assign organisation" link.

The welcome steps for new users now show your instance's name instead of an old "avodev" label.

Change requests now have the same name everywhere in the app, in every language.

In German the menu entry reads "Änderungsanfragen" instead of the ambiguous "Anfragen". The Organizations and Projects pages now open with one line explaining what they hold.

Contributors on a project now see the "New request" button, and admins can invite people from a project's Members page.

Before, only owners and maintainers saw "New request", although contributors were allowed to submit change requests. Admins who were not members of a project could not invite its team.

An instance deploy now fails when its GitHub environment has no Cloudflare credentials, instead of reporting success without uploading the frontend.Self-hosting

Before, the run warned, skipped the Pages upload and finished green, so the console recorded the new release while the old frontend stayed live. If you deploy the frontend yourself, dispatch with `deploy_pages=false`.

The manual-rework image (browser VS Code on a project's server) is now ghcr.io/product-printer/pp-rework: scanned, signed and 33% smaller.Self-hosting

Every build is checked for known vulnerabilities before it is published, signed with a keyless cosign signature, and ships its SBOM and build provenance. It contains no credential and needs none to build.

A proven requirement no longer turns red when a test run's results can't be found. That run now records nothing, and its CI check shows why.

A workstep whose build failed can now be run again with "Execute step" (or Retry in the pipeline), and a refused run says why.

A retry continues on the same branch and pull request. Steps that were queued behind a run that never started go back to pending, showing the reason.

Traceability now says why a test result doesn't count as proof, for example because the check was cancelled or the change was merged outside the gate.

A workstep's page now also follows a running step: it shows the step's progress and its pull request as soon as it opens, without reloading.

Answers you type into a proposed solution design's decisions are no longer lost when you reload the page.

They are kept in this browser until you commit the design.

Signed-in pages no longer log React hydration errors on load, and they render once instead of twice.

Requirements extracted with "Refine with agent" are no longer lost when you close the drawer or reload the page.

They stay until you add them or click Discard. They are kept in this browser only.

The assistant can now write a requirement for you: it opens the New requirement form and fills it in, and the chat stays usable while the form is open.

Check the filled fields and click Save yourself.

Traceability now counts a test result as proof only when the app's CI check completed successfully, never from a check that was cancelled or failed.

A change request now moves on by itself — in progress, merged, released in a version, deployed — so the client sees where their change is.

"Propose a design" no longer fails with "direction \"internal\" must be inbound, outbound or bidirectional" on small projects.

"New workstep" creates the workstep again instead of failing when it is added at the end.

Your app now has versions: see each release (v1.0, v1.1 …) and what changed in it under Releases.

A platform admin releases a new version from the same page: the version and the notes are proposed from the merged worksteps and change requests since the last release, and can be edited before publishing. It tags the app's repository and creates a GitHub Release.

A requirement now shows the pull request that delivered it, and the version it was released in, as soon as its workstep's PR is merged.

Traceability counted only PRs whose description named the requirement, so PRs built by the agents stayed at 0.

Traceability now shows, for every acceptance criterion, whether its automated test passed, with a link to the test run — and your client can see it too.

Apps created from now on report their test results after every run on the main branch. A developer maps each test to its criterion in `.avodev/requirement-tests.json`; criteria without a test show "No test yet".

New apps start with a green test check: test coverage can no longer drop below the level the app already has, and that level only goes up as tests are added.

After your first week, Product Printer asks one quick question: how likely you are to keep using it, and what's missing.

It appears once per person, bottom right. "Not now" hides it for three days; your answer goes to the Product Printer team without your name.

Acceptance criteria now show up in Traceability and work planning for every requirement, including those added from an extraction or edited by hand.

Requirements created before this update are fixed automatically.

Your instance's public pages now show only its own name, sign-in and legal links, with no marketing text from another company.

This covers the header, the footer, the page-not-found screen and the page titles search engines see.

Generated worksteps give each feature requirement to one step, and any conflict is shown right away instead of after the first change in chat.

Submitting a change request from a project works again instead of failing with a permission error.

Generated worksteps can be committed again when two requirements share acceptance-criterion numbers (AC-01, AC-02…).

Every app you create now checks each change for leaked secrets, vulnerable packages and unprotected data before it can merge.

The checks run on every pull request of the app's repository and fail it on a high-severity finding: a committed secret, a package with a known high-severity vulnerability, a database table without access rules, or a backend function anyone can call without signing in.

Creating a project from "New project" works again instead of failing with "permission denied".

You can accept, reject or defer a proposed requirement from its detail panel, which unblocks the solution design and worksteps.

Your instance no longer needs any AI provider key: models, embeddings and website imports use the keys on your orchestrator.Self-hosting

Only voice still uses a key on the instance, and you can switch it off in Admin → AI models.

Change requests filed as GitHub issues are imported again when your orchestrator runs without the instance's master database key.Self-hosting

The instance checks every imported request against its project's registered repository before saving it.

Your orchestrator now runs build steps and UI designs without your instance's master database key.Self-hosting

The instance gives each job what it needs and records the results itself; design files are uploaded through the instance for that job only, while it runs.

You can now talk with the assistant in the side panel, and it can fill in forms on the page for you to review.

Start voice mode from the side panel to speak and hear the replies. When the assistant fills a field, the field is marked and you can undo it. Take over whenever you want. Nothing is submitted until you click.

New workflow `instance-export.yml`: a full export of your instance (database dump, file listing, restore notes) as a download from the run page.Self-hosting

It uses the same GitHub environment as `build-and-deploy-instance.yml` and only reads your instance.

Members of a managed-cloud instance can now find it from product-printer.ai: they enter their e-mail address and receive a message listing the instances they belong to.Self-hosting

The new ops-directory function only answers a caller holding the dedicated INSTANCE_OPS_READ_SECRET. It returns the instance name, its address and one keyed hash per confirmed member (HMAC-SHA256 with a key derived from that secret), never an e-mail address. The link in that message opens the sign-in page with the address already filled in (`/login?email=…`).

Instances can now report a usage summary (member count, AI spend this month, last activity, release) to the Product Printer operator console.Self-hosting

The new ops-usage function only answers a caller holding the dedicated INSTANCE_OPS_READ_SECRET; it returns counts and totals, never user data. The deploy workflow now also records the deployed release in INSTANCE_RELEASE.

New change requests are drafted with the assistant in the side panel instead of a separate chat on the page.

Click "Draft with the assistant": the side panel opens on this change request with a starter message. The assistant fills the form, you check it and submit.

Everyone on a project can now open its Members page and see who is on the team, and My Projects is one click away from the header and your account page.

Inviting, changing roles and removing people stays with project owners and maintainers.

The repository now builds only the instance itself; the marketing sites are separate projects.Self-hosting

A plain `npm run build` produces the instance with its security headers in `dist/`, whatever variables are set. Old marketing paths (/pricing, /blog, …) redirect to `VITE_INSTANCE_MARKETING_URL` when it is set.

Choose how your instance sends email, and every email carries your brand.Self-hosting

Settings → Email offers your own Resend account or your own SMTP server (cloud adds a zero-setup Standard option). It checks the key, the sender's domain and the SMTP login live, lists DNS records Resend still needs and shows the provider's own error text. "Send test email to me" confirms delivery. Keys and passwords stay in your instance's Vault. Sign-up, password-reset, magic-link and invite emails carry your instance's brand, and every deploy keeps the 13 authentication templates in sync without overwriting the email provider you chose.

A new instance now looks like Product Printer until you brand it, and you can set its colours, corner radius, fonts and favicon in the instance config.Self-hosting

The same colours and font carry into the sign-in and notification emails. Button text on your primary colour switches between white and dark automatically so it stays readable.

When a deployment ships your change request, it now moves to Deployed and you get the "your change is live" email.

A missing field made the deployment step skip change requests silently. Prices are not affected.

The invitation email now takes you straight to choosing your password, then into the instance, instead of the sign-up form.

If the link has expired or was already used, the page says so and can email you a new one. Inviting someone who was already added to a project no longer fails.

Your answers to open questions now turn into proposed changes to your requirements, which you review and accept before anything changes.

The chat now lives in one panel at the side of every project page, with voice dictation, and remembers the conversation.

It replaces the separate chat windows that used to open on some pages.

A new Documents menu keeps the files of a project in one place, and the assistant now uses them when it answers.

The Product Printer website has a new front page that shows, step by step, how an order turns into a running app.

A new solution design step sits between your requirements and the work plan, so you can review how the app will be built before work starts.

A new "What's new" entry in your account menu shows what changed in each update, with a dot on your avatar when there is something you haven't seen yet.

A new sign-in page lets you find your workspace by its address and sign in there.

An instance can show its own announcement banner at the top of the app, set with `VITE_INSTANCE_BANNER` in the instance env; it is off by default.Self-hosting

The beta banner at the top of the app is gone.

The Product Printer blog has been retired: it is no longer linked from the website's menus.

After you sign in, you land where you work: straight in your project if you have one, otherwise on your project list.

Links you open while signed out still take you to the right page once you have signed in.

Long analyses now run in the background and show their progress, so you can keep working and never lose the result to a timeout.

If you have set up two-factor authentication, it is now asked for before any page of the app opens, not only on some of them.