Deployment Failure - Chat Fix Not Deployed

Date: January 1, 2026
Issue: Chat is still broken because the deployment failed
Status: Deployment failure prevented code fix from being deployed

Current Status

  • ✅ Code fix committed: GPT-5.1-chat API parameters fix is in the codebase
  • ❌ Deployment failed: Latest deployment (2026-01-01T00:01:39Z) failed
  • ❌ Chat still broken: Fix hasn’t been deployed to Azure Container Apps
  • ✅ Episodes/Sessions/Voice: Working (using older code without the fix)

Deployment History 

20629370015 | completed | failure | Deploy | 2026-01-01T00:01:39Z  ← Latest (FAILED)
20629361309 | completed | success | Deploy | 2026-01-01T00:01:06Z  ← Previous (SUCCESS)
20629291992 | completed | failure | Deploy | 2025-12-31T23:53:58Z  ← Earlier (FAILED)

Next Steps

Step 1: Check Deployment Failure Reason 

gh run view 20629370015 --log-failed

Common causes:

  • Build errors
  • Test failures
  • Azure deployment errors
  • Configuration issues
  • Resource constraints

Step 2: Fix Deployment Issue

Once the deployment failure reason is identified, fix it and re-deploy.

Step 3: Verify Deployment Succeeds 

gh run list --workflow=deploy.yml --limit 1

Wait for deployment to complete successfully.

Step 4: Test Chat Endpoint

After deployment succeeds:

python3 scripts/test-chat-debug.py \
  --token "YOUR_JWT_TOKEN" \
  --message "hi"

Expected: HTTP 200 with actual agent response (not error message)

Important Note: Endpoint Format Configuration

There’s a configuration discrepancy:

The documentation says:

  • AZURE_AI_ENDPOINT should include /openai/v1/ for OpenAI-compatible format

But the actual endpoint you’re using is:

  • https://zimax-gw.azure-api.net/zimax/openai/deployments/gpt-5.1-chat/chat/completions

This is the Azure Foundry format (not OpenAI-compatible format).

The code handles both formats:

  • ✅ If endpoint contains /openai/v1 → Uses OpenAI-compatible format (model in body)
  • ✅ Otherwise → Uses Azure Foundry format (deployment in path) ← This is what you’re using

For Azure Foundry format (your case):

  • Endpoint should be: https://zimax-gw.azure-api.net/zimax (base URL)
  • Code constructs: {base}/openai/deployments/{deployment}/chat/completions?api-version={api_version}
  • Our fix applies: max_completion_tokens and no temperature for gpt-5.1-chat

Configuration for Azure Foundry Format

Based on your actual endpoint usage:

AZURE_AI_ENDPOINT="https://zimax-gw.azure-api.net/zimax"  # Base URL (no /openai/v1/)
AZURE_AI_DEPLOYMENT="gpt-5.1-chat"
AZURE_AI_API_VERSION="2024-12-01-preview"
AZURE_AI_MODEL_ROUTER=""  # Empty = use direct model
AZURE_AI_KEY="cf23c3ed0f9d420dbd02c1e95a5b5bb3"

Code will construct:

https://zimax-gw.azure-api.net/zimax/openai/deployments/gpt-5.1-chat/chat/completions?api-version=2024-12-01-preview

Payload will include:

  • max_completion_tokens (not max_tokens) ✅ Fixed
  • No temperature parameter ✅ Fixed

Summary

  1. Code fix is correct - Handles Azure Foundry format with gpt-5.1-chat parameters
  2. Deployment failed - Fix hasn’t been deployed yet
  3. Need to fix deployment - Check failure logs and resolve
  4. Then re-deploy - Once fixed, deployment will include the chat fix
  5. Chat should work - After successful deployment with the fix
  • docs/troubleshooting/gpt-5.1-chat-api-parameters-fix.md - The code fix
  • docs/troubleshooting/chat-still-broken-after-parameter-fix.md - Troubleshooting guide
  • backend/agents/base.py - Code implementation