Unable to add azure-openai existing deployed models

[leonavevor]

Bug Description
I have created a new azure-openai (openai compatible) provider in the provider page

I tried to add a new model to my newly created agent in the model selection dropdown in the Agent configuration and added the actual model ID (e.g. gpt-5.1, gpt-5.2-codex) but when i press check it say model doesn't exist which is not true.

This makes it impossible for users to use models when using OpenAI-compatible providers like "azure-openai".

I was successful in a making it work with my local "ollama" without issues (just a bit of tweak to the docker compose to allow comm with the host network).

Root Cause:

Fix:
Remove the logic for checking or give possibility to optionally skip checking so we are not blocked on a false negative error that prevents us from saving the agent config.

[leonavevor]

i have managed to fix the issue and probably not a direct bug from goClow.
i had to go via a proxy like litellm to get it working but not directly to Azure-openai (it doesn't work as expected)

Remedy:

setup proxy via docker compose service

  # litellm
  # ui = http://localhost:4000/ui 
  # endopoint = http://localhost:4000

  litellm:
    image: litellm/litellm:v1.82.dev1
    network_mode: "host"
    #ports:
      #- "4000:4000"
    environment:
      # azure openai
      - AZURE_API_TYPE=azure
      - AZURE_API_KEY=${LITELLM_AZURE_API_KEY:-}
      - AZURE_API_BASE=${LITELLM_AZURE_BASE_URL:-}
      - LITELLM_MASTER_KEY=${LITELLM_MASTER_KEY:-}
      - UI_USERNAME=${UI_USERNAME:-}
      - UI_PASSWORD=${UI_PASSWORD:-}
      - DISABLE_ADMIN_UI=${DISABLE_ADMIN_UI:-}
      - ROOT_REDIRECT_URL=${ROOT_REDIRECT_URL:-}
    volumes:
      - ./litellm-config.yaml:/app/config.yaml 
    command: [ "--temperature", "1", "--config", "/app/config.yaml", "--port", "4000", "--detailed_debug"]

create litellm-config.yaml

model_list:
  - model_name: azure-gpt-5.3-chat
    litellm_params:
      api_base: os.environ/LITELLM_AZURE_BASE_URL
      api_key: os.environ/LITELLM_AZURE_API_KEY # The `os.environ/` prefix tells litellm to read this from the env.
      model: azure/gpt-5.3-chat # for older versions of azure ai studio, use "azure_ai/gpt-5.3-chat" 
      api_version: "2025-04-01-preview"
      #drop_params: true # drop all other params except these when using this model
      additional_drop_params: ["context_management", "output_config"] # or drop all params by setting drop_params: true (use-case: antropic claude sends these params but azure openai based doesn't support them, so we want to drop them when routing to azure-gpt-5.3-chat)
    # model_info:
    #   id: string
    #   mode: embedding
    #   input_cost_per_token: 0
    #   output_cost_per_token: 0
    #   max_tokens: 2048
    #   base_model: gpt-5.3-chat
    #   additionalProp1: {}

create .env (will be auto injected on docker compose up)

## Litellm
LITELLM_AZURE_BASE_URL=https://<project-name.here>.openai.azure.com
LITELLM_AZURE_API_KEY=***
## Litellm-ui
DISABLE_ADMIN_UI=false
ROOT_REDIRECT_URL="/ui"
LITELLM_MASTER_KEY=*** # used via proxy. usage: curl -H "Authorization: Bearer litellm" http://localhost:4000
UI_USERNAME=admin   # username to sign in on UI
UI_PASSWORD=***        # password to sign in on UI

proceed to run: docker compose up and add a new provider called "litellm_local" then configure you agent to use it and save (in the goClaw ui)

#Troubleshooting tips:
#1. curl http://127.0.0.1:4000/v1/models -H "Authorization: Bearer litellm" 
#2. curl --location 'http://0.0.0.0:4000/chat/completions' \                                                                                                                                                                                                                                                                                                                                                                                                                  ✘ INT 03:37:14 
#     --header 'Content-Type: application/json' --header 'Authorization: Bearer litellm' \
#     --data '{
#     "model": "azure-gpt-5.3-chat",
#     "messages": [
#         {
#         "role": "user",
#         "content": "what llm are you"
#         }
#     ]
# }'

enjoy!!!

[leonavevor]

maybe it will be worth integrating litellm-proxy into goClaw in future releases

[truongqv12]

The quickest workaround you can use right now is to create an external Docker network (e.g., shared-net) and attach goclaw to it.

You can do this by updating your compose file:

docker-compose.network.yml

services:
  goclaw:
    networks:
      - default
      - shared-net

networks:
  shared-net:
    external: true

After that, you just need to connect your litellm container to the same network. Once connected, instead of using localhost:4000, you can simply route to it using litellm:4000.