Is your feature request related to a problem? Please describe.
We need more robust, region-diverse, and cost-effective LLM options in Haystack pipelines. Relying on a small set of providers can cause issues during rate limiting, regional outages, or price-sensitive workloads. Adding CometAPI as an OpenAI-compatible provider would improve availability, portability, and resilience without disrupting existing workflows.

Describe the solution you'd like
Add CometAPI as a first-class LLM provider in Haystack, or officially support configuring the existing OpenAI components via base_url to point to CometAPI. Desired capabilities:

• Chat and text generation via OpenAI-compatible endpoints
• Model listing and selection via CometAPI model APIs
• API Key configuration and credential management
• Streaming and non-streaming support (leveraging Haystack abstractions if available)
• Minimal example + docs showing how to use CometAPI in a Haystack pipeline

Describe alternatives you've considered

• Use current OpenAI adapters with a custom base_url (if already supported). This works but lacks discoverability and may have subtle compatibility gaps.
• Go through third-party proxies/aggregators. This adds an extra dependency and less control.
• Stay with current providers only. This limits resilience, vendor flexibility, and cost optimization.

Additional context
CometAPI Resources:

• Website: https://www.cometapi.com/
• API Documentation: https://api.cometapi.com/doc
• Base URL: https://api.cometapi.com/v1/
• Model List: https://api.cometapi.com/v1/models
• Get API Key: https://api.cometapi.com/console/token
• Pricing: https://api.cometapi.com/pricing

Compatibility:

• CometAPI exposes OpenAI-compatible endpoints, enabling low-friction integration by reusing Haystack's existing OpenAI components where possible.

Implementation Offer:

• We’re happy to implement this integration and open a PR following your contribution guidelines (including examples and docs). We can also provide a temporary test API key and a small demo notebook/script if helpful. Please advise on the preferred integration point (new provider vs. extending OpenAI components) and acceptance criteria.