Skip to content

Bug: using Azure Foundry (OpenAI compatible endpoint) doesn't work #1566

Description

@drdr-kr

I tried using Azure Foundry as the model provider, my netclaw.json file looks like

{
  "Daemon": {
    "Host": "127.0.0.1",
    "Port": 5199,
    "ExposureMode": "local"
  },
  "Providers": {
    "azure-foundry": {
      "Type": "openai-compatible",
      "Endpoint": "https://my-foundry.cognitiveservices.azure.com/openai/v1"
    }
  },
  "Models": {
    "Main": {
      "Provider": "azure-foundry",
      "ModelId": "my-model",
      "ContextWindow": 128000
    }
  }
}

however this doesn't work as it seems like Netclaws OpenAiCompatibleChatClient isn't quite compatible with Azure Foundry.

Here's what Claude had to say:

src/Netclaw.Providers/SelfHosted/OpenAiCompatibleChatClient.cs:221 adds body["return_progress"] = true on every streaming request. The comment assumes unknown fields are ignored — true for llama.cpp (where the field comes from), false for Azure OpenAI, which rejects unknown args with a 400. The backend here is generic-openai, so the field should not be sent at all.

Metadata

Metadata

Assignees

No one assigned

    Labels

    providersProvider integrations and capability detection across OpenAI-compatible backends.

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions