You can install Dyad 1.11 as a desktop app on Windows, macOS or Linux, install Node.js when its setup banner asks, and connect an AI provider or a local model. Create a small app, approve the proposed code and use the preview panel to check it.
The useful new step in version 1.11 is test recording. Start a recording from the preview, complete a real user flow, and Dyad captures navigation, clicks and form input. After you stop, Dyad attaches a test-proposal request to the chat input. Sending that request in Agent mode produces reviewable steps and assertions; approving the proposal creates the Playwright test file.
Documentation-verified scope: I rechecked this guide on 19 August 2026 against Dyad’s 1.11.0 release notes and tagged source, quickstart, model, preview, debugging, privacy, versioning and publishing documentation. I did not install Dyad, generate an app, run its proposed test or deploy a project. The steps are documentation-verified rather than a claim of hands-on testing.
What changed in Dyad 1.11?
Dyad 1.11.0 was released on 18 August 2026. Its main addition is a recorder inside the app preview. Instead of asking an AI to guess how a feature should be tested, you perform the workflow and let Dyad capture the sequence. The resulting steps and assertions remain reviewable before approval creates the Playwright test.
The release also adds Auto Sidekick, sticky model selection for each chat, per-model reasoning-effort preferences and restoration of queued prompts after a restart. Auto Sidekick currently uses GPT-5.6 Sol to supervise a task while delegating scoped work to GPT-5.6 Luna. Delegated activity stays visible and existing approval and tool controls still apply. Dyad 1.11’s tagged model picker exposes Auto Sidekick to Dyad Pro users, and the model pairing can change, so treat it as a release-specific option rather than a permanent or free feature.
My recommendation is to use the recorder for one short, deterministic path first, such as adding and deleting a task. A generated test is only useful when its assertions describe the outcome you actually care about.
What you need
- A Windows, macOS or 64-bit Linux computer. The official 1.11.0 GitHub release provides a Windows setup file, Intel and Apple-silicon macOS archives, and Linux AppImage, Debian and RPM packages.
- The official Dyad download rather than a third-party mirror.
- Node.js, which Dyad uses to run generated JavaScript apps locally.
- AI access through a configured provider, Dyad Pro or a supported local-model setup. Creating the proposal uses Agent mode, so the selected model and any account quota must support that workflow.
- Internet access for the initial download, provider calls and npm packages unless your chosen local workflow already has everything cached.
- A throwaway test account and synthetic form data if the app includes authentication or personal information.
Privacy distinction: Dyad and the generated project run on your computer, but that does not make every AI request local. Dyad’s privacy policy says free-desktop prompts and code go directly to the model provider when you use your own key, while a supported local model keeps that material on the device. Dyad Pro requests are proxied through Dyad’s servers, where portions may be logged temporarily for abuse investigation or service reliability. Optional pseudonymous desktop telemetry is processed by PostHog and, according to the policy, excludes chats, prompts and code. Check the selected provider’s terms and never include secrets in a prompt or recording.
Step 1: download and open Dyad
Go to the official Dyad website or the project’s 1.11.0 GitHub release and use the download for your operating system. Windows uses the setup file, macOS has separate Intel and Apple-silicon archives, and 64-bit Linux has AppImage, Debian and RPM packages. Open or install the package using the normal process for that format.
Windows can display a Microsoft Defender SmartScreen warning. Dyad’s official quickstart says to select More info and then Run anyway. Only do this after confirming that the file came from the official Dyad site or release. A warning is not a reason to run a similarly named file from a mirror.
If you want an additional integrity check before bypassing an operating-system warning, calculate the file’s SHA-256 hash and compare it with the digest shown for that asset on the official GitHub release. Use Get-FileHash in PowerShell, shasum -a 256 on macOS or sha256sum on Linux. Do not disable platform security globally to make one application run.
Step 2: complete the Node.js setup
Dyad shows a home-screen banner for incomplete setup tasks. Select the Node.js task and follow its link to the correct installer for your computer. After installation, restart Dyad so it can detect the new runtime.
You can also open a terminal and check that Node.js is on your path:
node --version
npm --versionDyad tries to install pnpm because it can reuse dependencies across projects. If that fails, the app falls back to npm. The official Node.js troubleshooting page gives this manual pnpm command:
npm install -g pnpm@latest-10Restart Dyad after running it. Do not install pnpm manually unless Dyad reports that its normal setup failed.
Step 3: configure AI access
Open the remaining AI setup task from Dyad’s home-screen banner. Add the credential for the provider you intend to use, or configure a supported local model. The official model guide covers Gemini, OpenRouter, Anthropic, OpenAI and local models.
Start with the automatic model choice if you do not have a strong preference. Switch models when a result stalls rather than repeatedly sending the same vague request. Provider limits, billing and data policies belong to the provider account, even though you enter the key in Dyad.
Store an API key only in the intended credential field. Never paste it into the app’s source code, a chat prompt, a screenshot or a public Git repository.
Step 4: create a small first app
Enter a narrow request in the home-screen chat. For example:
Build a simple task list with an Add task form, an empty-state message,
checkboxes for completed tasks, and a Delete button. Keep data in local
browser storage. Do not add authentication or a cloud database.This scope is deliberate. It gives the recorder a form, a visible state change and a delete action without introducing credentials or an external service.
When Dyad finishes proposing code, inspect the change and approve it only if it matches the request. Approved code is saved on your computer and the app opens in the preview panel. Add a task, reload the preview and confirm that browser storage behaves as expected before recording anything.
If you are comparing this workflow with older repository-to-agent tools, the site’s GPT Engineer guide and GPT Pilot guide cover different command-line and coding-agent approaches.
Step 5: record one preview workflow
- Return the task list to a known starting state.
- Use the recording control in the app preview to start a test recording.
- Type a synthetic task such as
Check backupsinto the form. - Select the Add button.
- Confirm visually that the new task appears, then mark it complete.
- Stop the recording.
- Select Generate test proposal in the recorder bar.
According to the 1.11 release notes, Dyad captures actions such as navigation, clicks and form input. Stopping parks those actions as a draft; it does not create a test file. The proposal request is attached to the chat input, where you can refine its scope before sending it. If the current chat is not using Agent mode, Dyad asks permission to switch because the proposal stage needs to inspect the app and prepare the test.
Step 6: review the proposed assertions
Review the request in the chat input before sending it. Dyad’s Agent mode then proposes a test name, a plain-English description of the recorded steps and assertions in a review card. A useful test for this example should verify an observable outcome, not just repeat clicks. Check that it covers the state you care about:
- the newly entered task becomes visible;
- the completion control changes the task’s state;
- the test does not depend on a random delay or an unrelated page element;
- no real email address, password, API key or personal record appears in the captured input;
- the assertions are specific enough to fail when the feature breaks.
Edit, remove or reorder an assertion in the review card when it is vague, unsupported or misplaced. Approve the card only when the name, steps and checks match the flow; approval is what creates the test file. Then inspect the generated file and run it from the Tests panel. Dyad 1.11 also introduced more stable selectors for recorded tests, but stable selectors do not make a logically weak assertion correct.
Step 7: rerun the test after a change
Make one intentional change and run the test again. If it fails because a locator encoded text or structure you changed, decide whether the failure represents a real user-facing change or an unnecessarily fragile selector. Do not weaken an assertion merely to turn the result green.
If the change breaks the app, open the version control at the top of the chat panel. Dyad creates a version each time it edits code. Select the last good version and choose Restore. The versioning guide explains that restoration is additive, so the earlier history remains available.
Optional: sync and publish the app
You do not need to publish an app to test it locally. If the project is ready to share, Dyad’s publishing guide uses this flow:
- Create a GitHub repository. Choose private unless the source is intentionally public.
- Open the app details in Dyad and select Connect to GitHub.
- Use Sync to GitHub to push the project.
- Import that repository into Vercel or another compatible host.
- Add required environment variables in the host rather than committing them.
- Test the public URL and its access controls before sharing it.
Publishing changes the risk. A harmless local prototype can expose data, storage rules or unauthenticated actions when placed on the public internet. Test authentication and server-side authorization separately from a recorded happy-path UI flow.
Troubleshooting Dyad 1.11
Dyad still says Node.js is missing
Restart Dyad after the Node.js installer finishes, then confirm that node --version works in a new terminal. If it does not, repair the Node installation or its path. If the command works but Dyad still cannot detect it, use Dyad’s Help button and Report a bug; the official help flow pre-populates system information and recent logs.
The preview is blank or keeps loading
Open System Messages in the drawer below the preview. The first preview can take longer because Dyad installs npm packages. Use Restart when the Node.js server is stuck; use Refresh when only the page needs reloading. Copy a specific error into the chat instead of asking it to “fix the broken app.”
The provider returns a rate-limit error
Wait for the provider window to reset, reduce the material sent in one request, switch to another configured provider or enable billing after setting a spending limit. A model’s context window is not the same thing as an account’s requests-per-minute or tokens-per-minute quota.
The generated test passes without checking the result
Edit the proposal so it asserts the visible end state. Record one job at a time, begin from a known state and avoid workflows that rely on random content. A replay that only clicks buttons is not a meaningful regression test.
The recorder captured sensitive text
Do not send or approve the proposal. Remove the sensitive value, rotate it if it was a real credential and repeat the recording with synthetic data. Check the chat, generated files and version history before syncing the repository.
Frequently asked questions
Is Dyad completely local?
The desktop application and generated app run locally. AI inference is local only when you configure a local model. Free-desktop requests using your own key go directly to that provider, while Dyad Pro requests are proxied through Dyad’s infrastructure. Optional telemetry and user-initiated support uploads are separate data flows described in Dyad’s privacy policy.
Is Dyad open source?
Dyad uses mixed licensing. The repository states that code outside src/pro is open source under Apache 2.0. Code inside src/pro is fair source under the Functional Source License 1.1, with Apache 2.0 applying two years after each version is released. Check both licence files before reusing or redistributing the code.
Does Dyad 1.11 automatically add every recorded test?
No. The release notes say the test proposal is added to the chat input so you can review, refine or send it. Use that checkpoint; do not treat generated assertions as trusted code.
Do I need GitHub or Vercel?
No. They are part of Dyad’s documented publishing path, not prerequisites for building and testing a local app.
Will Auto Sidekick always be cheaper?
No fixed saving is guaranteed. Auto Sidekick is currently a Dyad Pro option that delegates straightforward work to a lower-cost model under a frontier model’s supervision. Actual credit use still depends on tokens, retries and task size, so check the usage shown in your Dyad account.
Bottom line
Install Dyad 1.11 if you want a desktop app builder with local project files and a review step before AI changes land. Its new recorder is most useful when you keep the first workflow small, inspect every proposed assertion and rerun it after each meaningful change.
Use cloud models for convenience or a local model when data must stay on the machine, but do not confuse a local desktop interface with local inference. For larger projects, keep GitHub private by default, store secrets outside the repository and test authorization separately from the generated UI checks.