DEV Community

Cover image for How to Use OpenClaw with Azure Foundry OpenAI Using LiteLLM Proxy
GDS K S
GDS K S

Posted on

How to Use OpenClaw with Azure Foundry OpenAI Using LiteLLM Proxy

#azure #llm #ai #opensource

TL;DR:

OpenClaw doesn't natively support Azure OpenAI. This guide shows you how to use LiteLLM as a proxy to route OpenClaw requests to Azure OpenAI, letting you use your Azure credits ($150 free tier, startup credits, enterprise subscriptions).

The Problem

OpenClaw is an incredible open-source AI coding agent with 124k+ GitHub stars. But when you try to configure it with Azure OpenAI, you'll find... it's not supported.

┌─────────────────────────────────────────────────────────┐
│  OpenClaw Model Selection                               │
├─────────────────────────────────────────────────────────┤
│  ✓ OpenAI (gpt-4o, gpt-4o-mini)                        │
│  ✓ Anthropic (claude-3.5-sonnet, claude-3-opus)        │
│  ✓ Google (gemini-2.0-flash, gemini-1.5-pro)           │
│  ✓ DeepSeek, Groq, Mistral, xAI...                     │
│  ✗ Azure OpenAI  ← Not available!                      │
│  ✗ Azure AI Foundry                                     │
└─────────────────────────────────────────────────────────┘
Enter fullscreen mode Exit fullscreen mode

This is a problem if you have:

  • Azure Sponsorship credits ($5,000 for startups/MVPs)
  • MSDN/Visual Studio subscription ($150/month Azure credits)
  • Enterprise policies requiring Azure-hosted models
  • Regional compliance needs (Azure has 60+ regions)

I've raised a feature request (#6056) for native Azure support, but until then, here's the workaround.

The Solution: LiteLLM Proxy

LiteLLM is an OpenAI-compatible proxy that translates API calls to 100+ LLM providers, including Azure OpenAI.

┌──────────────┐      ┌──────────────┐      ┌──────────────────┐
│   OpenClaw   │ ──── │   LiteLLM    │ ──── │  Azure OpenAI    │
│   Agent      │      │   Proxy      │      │  (your credits)  │
└──────────────┘      └──────────────┘      └──────────────────┘
     :18789              :4000           eastus.api.cognitive...
        │                   │                      │
        └── OpenAI API ─────┘                      │
                            └── Azure API ─────────┘
Enter fullscreen mode Exit fullscreen mode

OpenClaw thinks it's talking to OpenAI. LiteLLM translates to Azure format. You use Azure credits.

Prerequisites

Requirement Details
Azure OpenAI resource With GPT-4o or GPT-4o-mini deployed
Server/VM Ubuntu 22.04 recommended
Node.js 22+ for OpenClaw
Python 3.9+ for LiteLLM

Get Your Azure OpenAI Details

From Azure Portal → Azure OpenAI → your resource:

Setting Example
Endpoint https://eastus.api.cognitive.microsoft.com
API Key EuRBGeGG... (Keys and Endpoint section)
Deployment name gpt4o-mini (Model deployments section)
API Version 2024-10-21

Step 1: Install LiteLLM 

# Create virtual environment
python3 -m venv ~/litellm-venv
source ~/litellm-venv/bin/activate

# Install LiteLLM
pip install 'litellm[proxy]'
Enter fullscreen mode Exit fullscreen mode

Step 2: Configure LiteLLM

Create ~/litellm_config.yaml:

model_list:
  - model_name: gpt-4o-mini
    litellm_params:
      model: azure/gpt4o-mini          # "azure/" prefix + deployment name
      api_base: https://eastus.api.cognitive.microsoft.com
      api_key: YOUR_AZURE_OPENAI_KEY   # Replace with your key
      api_version: "2024-10-21"

  - model_name: gpt-4o
    litellm_params:
      model: azure/gpt4o
      api_base: https://eastus.api.cognitive.microsoft.com
      api_key: YOUR_AZURE_OPENAI_KEY
      api_version: "2024-10-21"

general_settings:
  master_key: "sk-litellm-your-secret-key"  # Any string you choose

litellm_settings:
  drop_params: true  # CRITICAL: OpenClaw sends params Azure doesn't support
Enter fullscreen mode Exit fullscreen mode

Important: The drop_params: true setting is critical. OpenClaw sends parameters like store that Azure OpenAI rejects. This setting tells LiteLLM to silently drop unsupported parameters.

Step 3: Test LiteLLM 

# Start LiteLLM
litellm --config ~/litellm_config.yaml --port 4000

# In another terminal, test it
curl http://localhost:4000/v1/chat/completions \
  -H "Authorization: Bearer sk-litellm-your-secret-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4o-mini",
    "messages": [{"role": "user", "content": "Say hello!"}]
  }'
Enter fullscreen mode Exit fullscreen mode

You should get a response from Azure OpenAI.

Step 4: Install OpenClaw 

# Install pnpm if needed
curl -fsSL https://get.pnpm.io/install.sh | sh -

# Install OpenClaw
pnpm add -g @anthropic/openclaw

# Verify installation
openclaw --version
Enter fullscreen mode Exit fullscreen mode

Step 5: Configure OpenClaw

This is where it gets tricky. OpenClaw's custom provider config requires specific fields that aren't well-documented.

Edit ~/.openclaw/openclaw.json:

{
  "gateway": {
    "mode": "local",
    "port": 18789,
    "auth": {
      "token": "your-gateway-token"
    }
  },
  "agents": {
    "defaults": {
      "model": {
        "primary": "litellm/gpt-4o-mini"
      },
      "models": {
        "litellm/gpt-4o-mini": {
          "alias": "GPT-4o Mini (Azure)"
        }
      }
    }
  },
  "models": {
    "mode": "merge",
    "providers": {
      "litellm": {
        "baseUrl": "http://localhost:4000/v1",
        "apiKey": "sk-litellm-your-secret-key",
        "api": "openai-completions",
        "models": [
          {
            "id": "gpt-4o-mini",
            "name": "GPT-4o Mini (Azure)",
            "reasoning": false,
            "input": ["text", "image"],
            "contextWindow": 128000,
            "maxTokens": 16384
          },
          {
            "id": "gpt-4o",
            "name": "GPT-4o (Azure)",
            "reasoning": false,
            "input": ["text", "image"],
            "contextWindow": 128000,
            "maxTokens": 16384
          }
        ]
      }
    }
  }
}
Enter fullscreen mode Exit fullscreen mode

Critical Configuration Points

Field Value Why It Matters
models.mode "merge" Merges custom providers with built-in ones
providers.litellm.api "openai-completions" Tells OpenClaw which API format to use
agents.defaults.models Model alias object Required for model selection UI
model.primary "litellm/gpt-4o-mini" Provider prefix + model ID

Missing the api field causes a cryptic error: No API provider registered for api: undefined. I've raised issue #6054 to improve this error message.

Step 6: Set Environment Variables

Create ~/.openclaw/.env:

OPENAI_API_KEY=sk-litellm-your-secret-key
OPENAI_BASE_URL=http://localhost:4000/v1
OPENCLAW_GATEWAY_TOKEN=your-gateway-token
Enter fullscreen mode Exit fullscreen mode

Step 7: Run as Systemd Services (Production)

LiteLLM Service

Create /etc/systemd/system/litellm.service:

[Unit]
Description=LiteLLM Proxy for Azure OpenAI
After=network.target

[Service]
Type=simple
User=your-username
WorkingDirectory=/home/your-username
ExecStart=/home/your-username/litellm-venv/bin/litellm \
    --config /home/your-username/litellm_config.yaml \
    --port 4000
Restart=always
RestartSec=10

[Install]
WantedBy=multi-user.target
Enter fullscreen mode Exit fullscreen mode

OpenClaw Service

Create /etc/systemd/system/openclaw.service:

[Unit]
Description=OpenClaw Gateway
After=network.target litellm.service
Requires=litellm.service

[Service]
Type=simple
User=your-username
WorkingDirectory=/home/your-username/openclaw-workdir
EnvironmentFile=/home/your-username/.openclaw/.env
ExecStart=/home/your-username/.local/share/pnpm/openclaw gateway \
    --port 18789
Restart=always
RestartSec=10

[Install]
WantedBy=multi-user.target
Enter fullscreen mode Exit fullscreen mode

Enable and start:

sudo systemctl daemon-reload
sudo systemctl enable litellm openclaw
sudo systemctl start litellm openclaw
Enter fullscreen mode Exit fullscreen mode

Step 8: Verify It Works 

# Check services
sudo systemctl status litellm openclaw

# Check LiteLLM logs (should show 200 OK responses)
sudo journalctl -u litellm -f

# Open OpenClaw chat
openclaw chat
Enter fullscreen mode Exit fullscreen mode

You should see responses from GPT-4o-mini, powered by your Azure credits!

Troubleshooting

Error: "azure does not support parameters: ['store']"

Fix: Add drop_params: true to litellm_settings in your config.

litellm_settings:
  drop_params: true
Enter fullscreen mode Exit fullscreen mode

Error: "No API provider registered for api: undefined"

Fix: Add "api": "openai-completions" to your custom provider in openclaw.json.

Error: "401 Incorrect API key" pointing to api.openai.com

Fix: OpenClaw is ignoring your LiteLLM proxy. Ensure:

  1. Custom provider has baseUrl set correctly
  2. OPENAI_BASE_URL is set in environment
  3. Model reference uses provider prefix: litellm/gpt-4o-mini

Empty responses (agent completes but no output)

Fix: Usually the drop_params: true issue. Check LiteLLM logs:

sudo journalctl -u litellm -n 50
Enter fullscreen mode Exit fullscreen mode

Cost Comparison

Model OpenAI Direct Azure OpenAI
GPT-4o-mini input $0.15/1M $0.15/1M
GPT-4o-mini output $0.60/1M $0.60/1M
GPT-4o input $2.50/1M $2.50/1M
GPT-4o output $10.00/1M $10.00/1M

Same pricing, but Azure lets you use:

  • Sponsorship credits ($5,000 for startups)
  • MSDN credits ($150/month)
  • Enterprise agreements
  • Reserved capacity discounts

Conclusion

Until OpenClaw adds native Azure OpenAI support (vote on #6056), LiteLLM is the bridge you need. The setup takes about 15 minutes, and then you get the full OpenClaw experience powered by your Azure credits.

Architecture recap:

┌─────────────────────────────────────────────────────────────┐
│                      Your Infrastructure                     │
├─────────────────────────────────────────────────────────────┤
│                                                              │
│  ┌──────────┐    ┌──────────┐    ┌─────────────────────┐   │
│  │ OpenClaw │───▶│ LiteLLM  │───▶│    Azure OpenAI     │   │
│  │  :18789  │    │  :4000   │    │ eastus.cognitive... │   │
│  └──────────┘    └──────────┘    └─────────────────────┘   │
│       │                                     │               │
│       │         OpenAI-compatible           │               │
│       └────────────  API  ──────────────────┘               │
│                                                              │
│  Uses: $5k Azure Sponsorship / MSDN Credits / Enterprise   │
└─────────────────────────────────────────────────────────────┘
Enter fullscreen mode Exit fullscreen mode

Resources

Let's Discuss 💬

I'd love to hear from you:

  • Are you using Azure credits for AI development? What's your setup?
  • Have you tried other LLM proxies? How does LiteLLM compare?
  • What AI tools need Azure integration? Drop a comment!

Help the community:

  1. ⭐ Star OpenClaw on GitHub
  2. 👍 Upvote Issue #6056 for native Azure support
  3. 🔄 Share this guide with developers who have Azure credits

Keywords

Azure OpenAI · LiteLLM Proxy · OpenClaw Configuration · AI Coding Agent · GPT-4o Azure · Azure Credits · LLM Gateway · Open Source AI · MSDN Credits AI · Azure Sponsorship

Published by GDSKS • Lead Architect, Founder at GLINCKER • Building GLINR Platform

Found this helpful? Give it a ❤️ and follow for more Azure + AI tutorials!

Top comments (1)

Collapse
 
peacebinflow profile image
PEACEBINFLOW •

Appreciate everyone who’s been asking about Azure support in OpenClaw — this post is basically my “fine, I’ll just route around it” moment 😅

Quick recap: OpenClaw doesn’t speak Azure OpenAI (or Azure AI Foundry) natively yet, so the move is:

OpenClaw → LiteLLM Proxy → Azure OpenAI
OpenClaw thinks it’s talking to OpenAI, LiteLLM translates the requests, and you get to burn Azure credits instead of paying direct.

The two things that matter most (because they cause 90% of the pain):

drop_params: true in LiteLLM (Azure rejects some params OpenClaw sends, like store)

setting "api": "openai-completions" in OpenClaw’s provider config (otherwise you get the dumb “api: undefined” error)

If you’re using startup sponsorship credits / MSDN / enterprise Azure policies, this makes OpenClaw actually usable in those environments right now.

Curious though — who here is trying to run agents through Azure for compliance/region reasons vs just because free credits are free credits? 😂
Drop your setup + what you got stuck on, and I’ll help troubleshoot in the comments.