I am reporting an ongoing issue with npm package installation inside ChatGPT’s container execution environment.
The container is configured to use the following internal npm registry:
https://packages.applied-caas-gateway1.internal.api.openai.org/artifactory/api/npm/npm-public
This registry is automatically supplied through the container environment. I did not manually configure it.
Initial results
During repeated tests on July 24–25, 2026, the internal registry consistently returned:
HTTP 503 Service Temporarily Unavailable
npm error: E503
Server: nginx
The failure affected both:
npm ping
npm ci
Independent HTTP requests to the same registry also returned HTTP 503.
Current results
A new test on August 2, 2026 shows that the behavior has changed, but npm installation is still unavailable.
The internal registry now returns:
HTTP 404 Not Found
Server: nginx
The following checks all fail:
npm ping
npm view lodash version
Direct HTTP requests to the registry root and package metadata paths such as /lodash and /react also return HTTP 404.
The container reports:
Node.js: v22.16.0
npm: 10.9.2
NETWORK=caas_packages_only
Trying to temporarily use the public registry does not provide a workaround:
npm --registry=https://registry.npmjs.org/ ping
This fails because the container cannot directly resolve or access registry.npmjs.org.
Why this does not appear to be a project issue
The failure occurs before the project code is executed and can be reproduced with registry-level commands that do not depend on a specific repository or package-lock.json.
The same project and lockfile work correctly outside the affected ChatGPT container environment.
This does not appear to be caused by:
- incorrect use of
npm ci, - a malformed
package-lock.json, - a missing individual package,
- the project source code,
- or a browser-side networking issue.
Expected behavior
The internal npm proxy should provide access to npm package metadata and tarballs so that package installation can run inside the ChatGPT container.
Actual behavior
The configured internal npm route previously returned HTTP 503 and now returns HTTP 404 for all tested npm endpoints.
I am not claiming that every ChatGPT user or every container pool is affected. This may be limited to a specific container configuration, cluster, region, or rollout cohort.
Could OpenAI please verify whether the internal route below is currently configured and available?
/artifactory/api/npm/npm-public
It would also be helpful if other affected users could run the following read-only checks and share their results:
npm config get registry
npm ping --fetch-retries=0 --fetch-timeout=10000 --loglevel=verbose
node --version
npm --version
Please do not publish authentication tokens, passwords, or unredacted environment variables.