Added support for OpenRouter & Generic Models

[rglozman]

Added support for OpenRouter which hosts over 300 models. This includes support for models from every provider into one easy to use API.

I created an example using Gemini (free model hosted by google), updated the .env, and the README.

[chr-hertel]

Hi, thanks for the idea and I like the GenericModel approach for this platform 👍

Was just creating an API key for openrouter, was running the example you pushed and got this error:

php examples/chat-gemini-openrouter.php 
PHP Fatal error:  Uncaught PhpLlm\LlmChain\Exception\RuntimeException: Response does not contain message in /.../llm-chain/src/Bridge/OpenRouter/Client.php:54
Stack trace:
#0 /.../llm-chain/src/Model/Response/AsyncResponse.php(33): PhpLlm\LlmChain\Bridge\OpenRouter\Client->convert()
#1 /.../llm-chain/src/Chain.php(75): PhpLlm\LlmChain\Model\Response\AsyncResponse->unwrap()
#2 /.../llm-chain/examples/chat-gemini-openrouter.php(26): PhpLlm\LlmChain\Chain->call()
#3 {main}
  thrown in /home/christopher/Projects/LlmChain/llm-chain/src/Bridge/OpenRouter/Client.php on line 54

does it work for you?

---

btw, this should fix the codestyle issues in the pipeline:

PHP_CS_FIXER_IGNORE_ENV=1 vendor/bin/php-cs-fixer fix

or to run the entire quality pipeline locally:

make qa

[rglozman]

Could you add var_dump($data) just before line 54 in src/Bridge/OpenRouter/Client.php and run again? Inside the if statement just before the "throw new". Could be a bunch of different errors (auth, out of credits, servers down, etc...). I am working on a better way to report errors to the user (for all providers), I will make it a separate pull request.

Yes I will run the linter, thanks for the heads up.

[chr-hertel]

You are right, just tried it again with dump on the response. First time it worked, but second time it was rejected due to rate limit.

The error handling is something to improve in general. I would love to also have a more generalistic approach, but also the error responses depend on the specific model and platform. so a combination could be worth exploring 🤔 like a better default error handling and an extension point per platform/model that enables bridges to fine-tune specifics of error handling.

[chr-hertel]

So, all in all, thanks for the PR, happy to merge once the pipeline is green :) 👍

[rglozman]

Should be good to merge now 👍

[chr-hertel]

Thanks @rglozman! 👍