---
title: Troubleshooting
description: Fix a missing banner, analytics that load before consent, choices
  lost on reload, CORS errors, hydration differences, failed manifest downloads,
  mode errors and failed static builds in c15t v3.
group: guides
lastModified: "2026-10-10T16:01:45+01:00"
---
## Start with three checks

Most problems show up in one of these places. Run them before changing code.

1. **Which c15t is running.** In the browser console, run `window.c15t`. v3
   prints `{ version, pkg, mode, hosting }`, for example `mode: 'manifest'`
   from `@c15t/nextjs`. `hosting` is `'inth'` or `'self-hosted'` once the
   backend has answered `/init`, and `null` before then or without a backend. `undefined` means no c15t provider has mounted on this
   page. A `window.c15tStore` object means the page runs v2. With the script
   tag, `window.c15t` is the full browser API instead.
2. **The backend request.** In DevTools Network, filter by your backend URL.
   Look for `/init` or `/manifest` and check its status. A CORS error means
   the backend does not trust your site's origin.
3. **Consent state.** Add the c15t DevTools panel from your framework's guide.
   It shows the resolved policy, whether a prompt is needed, each category's
   permission and the scripts c15t manages.

Your framework's troubleshooting page covers adapter-specific failures, such
as Next.js prerendering errors or SvelteKit hydration.

## "Can't resolve 'c15t/next'" or a missing export

npm's `latest` tag still points to c15t v2, which has no framework entry
points. A plain `npm install c15t` therefore installs v2, and imports such as
`c15t/next`, `c15t/react` or `ConsentRoot` fail with "Package subpath is not
defined by exports" or "Module not found".

Install the v3 release and check the lockfile:

|Package manager|Command|
|:--|:--|
|npm|`npm install c15t@alpha`|
|pnpm|`pnpm add c15t@alpha`|
|yarn|`yarn add c15t@alpha`|
|bun|`bun add c15t@alpha`|

Keep every c15t package, including `@c15t/integrations`, on the same release.
Mixing v2 and v3 packages is not supported.

## Why is there no banner?

No banner is sometimes correct. Find out which case you have:

|What DevTools shows|Cause|Fix|
|--|--|--|
|The backend request failed or has a CORS error|Wrong backend URL, or your origin is not trusted|Copy the URL from your Inth project exactly, including any path. Add the site's origin, such as `https://www.example.com` or `http://localhost:3000`, to the project's trusted origins.|
|Requests go to `your-project.inth.app` and fail|The placeholder backend URL from the guides is still in your code|Replace `https://your-project.inth.app` with the URL from your Inth project, including any path prefix.|
|No backend request at all|No provider mounted, the provider runs in offline mode, or the page resolved from a bundled manifest|Check `window.c15t`. `mode: 'manifest'` with no request is expected for a policy that does not depend on location. Otherwise confirm your framework's backend URL variable is set, as [consent modes](/docs/concepts/modes#set-the-backend-url) lists.|
|The policy resolved and no prompt is needed|The visitor's location maps to a policy without a prompt, such as a US state without a privacy law in the recommended rules|Expected. Test from a location that needs a prompt. Server helpers such as Next.js `resolveConsent` accept a `country` override for testing; remove it before you deploy.|
|The policy resolved and a choice is stored|A returning visitor|Expected. Clear site data or use a private window.|
|The banner is in the DOM but invisible or unstyled|c15t's rules are missing: `styles: false` without an imported stylesheet, a Content Security Policy that blocks c15t's `<style>` elements, Tailwind CSS 3's preflight, or a SvelteKit app without `c15tHandle`|See [stylesheets and CSS layers](/docs/customization/stylesheets) for how your framework loads c15t's rules.|

Do not "fix" a missing banner by granting every category or turning c15t off.
While the policy is unresolved, every optional category stays denied. Turning
the runtime off lets optional scripts load.

## Why does analytics load before the visitor chooses?

c15t only controls scripts you register with it. Find every other loader:

* A `<script>` tag in your HTML, layout or `_document`.
* A framework package or plugin, such as `@next/third-parties`,
  `@nuxt/scripts`, `nuxt-gtag` or a Vercel or Netlify analytics toggle.
* A tag in Google Tag Manager that fires on page view.
* An embed, such as a YouTube iframe, rendered without `ConsentGate`.

Remove the extra loader and register the vendor through
[integrations](/docs/integrations/overview). To replace a framework package,
follow [migrate from `@next/third-parties`](/docs/frameworks/next/scripts#migrate-from-nextthird-parties),
[migrate from `@nuxt/scripts`](/docs/frameworks/nuxt/scripts#migrate-from-nuxtscripts)
or [migrate from `nuxt-gtag`](/docs/frameworks/nuxt/scripts#migrate-from-nuxt-gtag).
Then reload with the Network panel open and cache disabled. The vendor's
domain should be absent until you allow its category.

Also check the policy. Under an opt-out policy, optional categories are allowed
before a choice, so the request is expected.

Google Tag and Google Tag Manager are an exception by design. By default
their helpers load before a choice and send Google Consent Mode signals, so a
request to Google before consent is expected. If you need no request at all
before consent, set `loadMode: 'after-consent'`; see
[Google Tag Manager](/docs/integrations/google-tag-manager#choose-when-the-container-loads).

## Why does the choice disappear on reload?

|Check|Fix|
|--|--|
|After saving, does DevTools Application show a `c15t` cookie and a `c15t` localStorage entry?|If not, look for `persistence: false` in your config. Examples use it on purpose; production apps should not.|
|Does the browser block storage, for example in a sandboxed iframe or a strict privacy mode?|The choice works for the current page only. Nothing to fix in your app.|
|Did the domain, subdomain or protocol change between visits?|Cookies and localStorage are per origin. Serve the site from one origin, or share the cookie across subdomains.|
|Did the policy change since the visitor chose?|A changed policy, a new required category or an expired choice prompts again. This is intended.|

Never save a choice automatically on page load to make persistence "stick".
That records consent the visitor did not give.

## Why does the page reload after saving preferences?

A visitor turned off a category or vendor they had allowed. Scripts that
already ran cannot be unloaded, so c15t reloads the page to start clean with
only permitted code. Set `reloadOnConsentRevoked: false` if you clean up
revoked vendors yourself; [clear on revocation for your framework](/docs/integrations/overview#vendor-switches-and-cookie-cleanup)
covers the options.

## Why does the server HTML differ from the browser?

Pass the value your framework's server helper returns to the provider
unchanged. Do not merge it with defaults or rebuild it from permissions. Make
sure the server and browser use the same backend URL, because two backends can
resolve different policies.

Never keep consent state in a module-level variable on the server. One
visitor's state can leak into another request. Create it per request, as the
framework guides do.

## Why does the build fail to download the manifest?

`manifest()` resolves from a policy snapshot that the build downloads from
`${backendURL}/manifest`. A production build stops when the download fails,
and dev logs the same message as a warning. Every framework uses the same
messages, prefixed with the integration's name, such as
`@c15t/nextjs/build`, `@c15t/core/build` (for `c15t/build`),
`@c15t/tanstack-start/build`, `@c15t/vue/vite`, `@c15t/svelte/vite`,
`@c15t/vue` (Nuxt) or `@c15t/astro`:

|Message|Cause|Fix|
|--|--|--|
|`could not fetch the consent manifest from <url> during the build (fetch failed: ...)`, with `ENOTFOUND`, `ECONNREFUSED` or another network error|The build cannot reach the backend|Build where the backend is reachable, or deploy with `C15T_ON_BUILD_ERROR=runtime`|
|`... (no response within 10 seconds)`|The backend did not answer in time|Check the backend, or build where it is reachable|
|`... (/manifest responded 404 Not Found)`, or another status|The URL is not your project's backend, such as the `https://your-project.inth.app` placeholder or a URL missing its path prefix|Copy the backend URL exactly as Inth shows it|
|`... (/manifest returned an invalid consent manifest.)`|The URL answered with something other than a consent manifest, such as an HTML page|Point the backend URL at the backend itself, not your site or a dashboard page|
|`no backend URL is set, so the build cannot fetch the consent manifest. Pass backendURL or set ...`|No `backendURL` option and no backend URL variable|Set the variable your framework reads; see [consent modes](/docs/concepts/modes#set-the-backend-url)|
|`skipped the consent manifest fetch because "/api/c15t" is not an absolute http(s) URL, so the server fetches the policy at runtime.`|A relative backend URL. A notice, not an error.|Pass the absolute backend URL. With `onBuildError: 'fail'`, this stops the build with `build-time manifests require an absolute upstream URL.`|
|`C15T_ON_BUILD_ERROR must be 'runtime' or 'fail', received "..."`|The variable has another value|Set it to `runtime` or `fail`, or unset it|
|`onBuildError must be 'runtime' or 'fail', received "..."`|The option has another value|Pass `'runtime'` or `'fail'`|

Each failure message ends with what to do next. In a production build:
`` Set `C15T_ON_BUILD_ERROR=runtime` (or `onBuildError: 'runtime'`) to deploy with runtime fetching. ``
The Vite plugins add one more fix. The plugin can't see a mode set in app
code, so an app that uses `manifest({ manifestURL })` or
`manifest({ source: 'runtime' })` should pass `source: 'runtime'` to
`consentManifest()`, and the build skips the download.
A failed download never reuses an older snapshot. With
`C15T_ON_BUILD_ERROR=runtime`, the build finishes without a snapshot, and the
manifest is fetched at runtime. Dev warnings end with
`The server fetches it at runtime instead. A production build stops on this error.`

[Consent modes](/docs/concepts/modes#what-happens-when-the-download-fails)
describes the policy and when the build skips the download.

### A single-page app warns that the policy depends on location

`consentManifest()` from `c15t/build` logs this warning when it downloads a
policy with location rules:

```txt
@c15t/core/build: the consent policy depends on the visitor's location, which the browser doesn't know. Unless you pass `inputs` or `geoURL` to `manifest()`, every first visit still calls the backend's /init, so the bundled policy adds bytes without saving a request. Consider `mode: hosted()` instead.
```

Pass the visitor's `{ country, region }` as `manifest({ inputs })` when your
edge knows it, or a `geoURL` that answers with it. Otherwise switch the mode
to `hosted()`, which asks `/init` without bundling the policy.

## Why does the provider say a mode is required?

The browser throws when a provider or `init()` gets no mode, or a mode name
instead of a factory:

|Message|Fix|
|--|--|
|`` @c15t/react ConsentProvider: `mode` is required. Use manifest() or hosted(). ``|Pass `mode`, such as `manifest()` from the same package. The start of the message names the package and the API that got no mode, such as `@c15t/svelte ConsentProvider` or `@c15t/core createConsentRuntime()`. Production builds name the package only: `` @c15t/react: `mode` is required. ``|
|`` @c15t/browser: `mode` must be a factory, such as manifest(), hosted() or offline() from @c15t/browser. Mode names like "hosted" work only in the script-tag builds. ``|Import the factory and call it: `init({ mode: hosted() })`.|
|`` c15t: hosted() needs `backendURL`. Pass it, or add consentManifest() from c15t/build to your Vite config and set VITE_C15T_BACKEND_URL (or VITE_INTH_PROJECT_URL). ``|The React `hosted()` found no backend URL. `@c15t/browser` throws the same message prefixed `@c15t/browser:`. A production build prints only the first sentence.|
|`` c15t: `mode` is the hosted() transport itself, so its code ships in the client bundle. Pass the hosted() your framework package exports instead: it is plain data, and the root loads the code only when it runs. ``|A warning outside production. A server-rendered root got a transport from `c15t/react`. Import the mode from your framework package. Next.js has its own wording; see [Next.js troubleshooting](/docs/frameworks/next/troubleshooting).|

## Why does the static build fail when development works?

A static host serves files only. Route handlers, server functions, rewrites
and proxies that worked under the dev server do not exist after deployment.
Point the browser at the absolute backend URL from Inth instead of a
same-origin `/api/c15t` path, then test the built output with a static file
server, not the dev server. [Choose your setup](/docs/concepts/choose-your-setup)
lists the static path for each framework.

## Why does the UI disappear with an ad blocker?

Check the Network panel for `ERR_BLOCKED_BY_CLIENT`. Some filter lists block
the consent backend's domain or old c15t chunk names such as
`consent-dialog-*.js`. Current releases use neutral chunk names; update and
rebuild. If the backend domain is blocked, the banner cannot load policy and
every optional category stays denied.

## Why does my theme do nothing?

Check that c15t's rules load, that you set a token or slot the
component actually reads, and that the state attribute you target is on the
element you style. A `data-variant` on the banner root is not on its card.
Check the Tailwind version and CSS layer order before adding specificity. See
[customization](/docs/customization/overview).