AI & Agent Dev Bug Sandbox Back to Radar

HuggingFaceEndpoint Drops Token For Dedicated Inference Endpoints (*.Endpoints.Huggingface.Cloud)

The HuggingFaceEndpoint class discards the explicitly provided huggingfacehub_api_token when endpoint_url points to a dedicated Inference Endpoint served from *.endpoints.huggingface.cloud, because the internal _is_huggingface_hosted_url check does not recognise this hostname. This causes requests to protected endpoints to be sent without an Authorization header, resulting in 401 Unauthorized, and may silently replace the explicit token with a cached or environment token if present.

highConfidence 95%Langchain-Huggingface Affected V1.2.2

Origin Analysis

The helper _is_huggingface_hosted_url only treats huggingface.co and hf.space (and their subdomains) as Hugging Face hosted URLs. Dedicated Inference Endpoints use the domain *.endpoints.huggingface.cloud, which is not covered by the check, so validate_environment considers the endpoint as local and omits the token when instantiating InferenceClient and AsyncInferenceClient.

1. Install langchain-huggingface 1.2.2 and huggingface_hub 1.31.0. 2. Run the provided reproduction code that instantiates HuggingFaceEndpoint with endpoint_url='https://jzgu0buei5.us-east-1.aws.endpoints.huggingface.cloud' and an explicit huggingfacehub_api_token. 3. Observe that _is_huggingface_hosted_url returns False and both InferenceClient and AsyncInferenceClient are called with api_key=None.

Fixing Code Block

0x7F 45 4C 46 02 01 01 000x90 12 AB 04 C1 7E 33 08
0xA4 1C 90 3E 77 0B D2 580x11 F6 6A 2C 80 19 DE 44
0x00 00 00 00 03 00 3E 000x01 00 B7 4D 55 90 0C 61
0x6C 69 62 63 2E 73 6F 2E0x36 00 8E 17 42 AA 05 F3
0xDE C0 DE 01 8B 45 FC 890xC7 5D 20 6F 91 03 E4 78
0x13 37 00 4F 9A 2D E1 700x08 B4 5C 22 71 D0 4A 16
0x48 8D 3D 22 01 00 00 E80x91 FF 0A 63 2E 57 B8 04
0x5A 91 C3 0E 44 70 1B 880x02 6F 39 C4 15 80 ED 27
0xFF 15 40 20 60 00 48 890xC3 90 7A 01 4C 28 B1 56
0x0F 1F 44 00 00 55 48 890xE5 C3 18 6D 92 0B 33 A7

Edge Case Audit

The change is minimal and only broadens the set of hostnames considered Hugging Face hosted. However, it means the token will be sent to any subdomain of endpoints.huggingface.cloud; that domain is controlled by Hugging Face, but if it were ever compromised or repurposed, tokens could leak. No threading or concurrency issues are introduced because the helper is a pure function. Rollback is straightforward: revert the hosted_domains tuple to the previous two entries. For additional safety, users can still override headers via server_kwargs as a workaround.