The problem
Jitbit's AI features currently use the Chat Completions API. That's becoming a blocker with the newest OpenAI models:
- GPT-6 only allows function tools on Chat Completions with reasoning turned off. AI replies and tools (KB search, ticket history, escalation, MCP/external tools) keep working, but the model can't reason during tool use.
- GPT-6.1 Sol reportedly drops the "no reasoning" option and requires the Responses API for tool calling. Tool-based AI features wouldn't work with it at all.
- On Azure OpenAI, deployment names are chosen by the customer, so model-specific behavior shouldn't depend on the name.
The idea
Add a Responses API path for the OpenAI and Azure OpenAI providers, and keep Chat Completions for Gemini, Claude, Bedrock and OpenAI-compatible providers.
Nice to have:
- A stateless mode (no ticket content stored on the provider side), important for data retention and GDPR.
- Configurable reasoning effort in the AI settings.
- Reasoning tokens counted in the output limits and usage reports.