Bug type

Behavior bug (incorrect output/state without crash)

Summary

When we onboard openclaw to our machine through the onboarding wizard, if we try to configure an azure openai endpoint through the custom providers option, the onboarding is successful but tui, agent, etc does not work, it fails with 404 resource not found error.

Root Cause:
The issue happens because when an Azure OpenAI endpoint is onboarded through the custom provider option through the onboarding wizard, it is treated like a normal openai endpoint, specifically openai-completions. The issue here is that, pi-ai does not have any implementation for azure-openai-completions, it only has implementation for azure-openai-responses.

Suggested Fix
A good fix here would be that during onboarding, the azure openai endpoint details be stored as azure-openai-responses type instead of openai-completions. I would be more than happy to contribute and fix this issue once and for all.

Steps to reproduce

1. Install openclaw
2. Onboard using openclaw onboard
3. Select custom providers when it asks for provider details
4. Enter valid azure openai endpoint details
5. The onboarding is successful but if we "openclaw tui" or "openclaw agent...", it fails with 404 error

Expected behavior

openclaw tui and openclaw agent should work as expected

Actual behavior

"openclaw tui", "openclaw agent..." fails with 404 error

OpenClaw version

2026.3.13

Operating system

Ubuntu 24.04

Install method

pnpm dev

Model

gpt-5.2-chat

Provider / routing chain

openclaw -> azure openai -> gpt-5.2-chat

Config file / key location

No response

Additional provider/model setup details

Config example:

{
  "wizard": {
    "lastRunAt": "2026-03-15T06:13:42.390Z",
    "lastRunVersion": "2026.3.14",
    "lastRunCommand": "onboard",
    "lastRunMode": "local"
  },
  "models": {
    "mode": "merge",
    "providers": {
      "custom-someresourcexyz-openai-azure-com": {
        "baseUrl": "https://someresourcexyz.openai.azure.com/openai/deployments/gpt-5.2-chat",
        "apiKey": "someapikey",
        "api": "openai-completions",
        "models": [
          {
            "id": "gpt-5.2-chat",
            "name": "gpt-5.2-chat (Custom Provider)",
            "reasoning": false,
            "input": [
              "text"
            ],
            "cost": {
              "input": 0,
              "output": 0,
              "cacheRead": 0,
              "cacheWrite": 0
            },
            "contextWindow": 16000,
            "maxTokens": 4096
          }
        ]
      }
    }
  },
  "agents": {
    "defaults": {
      "model": {
        "primary": "custom-someresourcexyz-openai-azure-com/gpt-5.2-chat"
      },
      "models": {
        "custom-someresourcexyz-openai-azure-com/gpt-5.2-chat": {}
      },
      "workspace": "/root/.openclaw/workspace"
    }
  },
  "tools": {
    "profile": "coding"
  },
  "commands": {
    "native": "auto",
    "nativeSkills": "auto",
    "restart": true,
    "ownerDisplay": "raw"
  },
  "session": {
    "dmScope": "per-channel-peer"
  },
  "gateway": {
    "port": 18789,
    "mode": "local",
    "bind": "loopback",
    "auth": {
      "mode": "token",
      "token": "sometoken"
    },
    "tailscale": {
      "mode": "off",
      "resetOnExit": false
    },
    "nodes": {
      "denyCommands": [
        "camera.snap",
        "camera.clip",
        "screen.record",
        "contacts.add",
        "calendar.add",
        "reminders.add",
        "sms.send"
      ]
    }
  },
  "meta": {
    "lastTouchedVersion": "2026.3.14",
    "lastTouchedAt": "2026-03-15T06:13:42.400Z"
  }
}

Logs, screenshots, and evidence

Impact and severity

• Impacts azure subscription customers using azure openai
• Blocks workflow
• Frequency always
• Consequence is failed onboarding

Additional information

The issue happens because when an Azure OpenAI endpoint is onboarded as a custom provider through the onboarding wizard, it is treated like a normal openai endpoint, specifically openai-completions.
The issue here is that, pi-ai does not have any implementation for azure-openai-completions, it only has implementation for azure-openai-responses, hence, a good fix here would be that during onboarding, the azure openai ednpoint details be stored as azure-openai-responses type instead of openai-completions.

I would be more than happy to contribute and fix this issue once and for all.