Add support for LLM configuration inside the app

[Dobidop]

Problem

The application currently relies on hardcoded(?) endpoints for LLM and diffusion model sources, making it inflexible and limited in terms of backend model providers.

Issue

Users are unable to reconfigure the application to point to alternative LLM or diffusion model backends (e.g., local models or self-hosted APIs). This design prevents the use of local models and restricts the app to a single model provider, such as OpenAI.

Proposed Solution

Introduce a configurable setting inside the app that allows users to specify a custom API endpoint (compatible with OpenAI-like APIs). The application could query the /models endpoint (or equivalent) to dynamically fetch and list the available models. This would enable support for local models or other third-party APIs without code modification.

Image generation from alternative diffusion models (perhaps via a comfyui instance) is of less importance, but would be useful too.

[baturyilmaz]

Thank you for bringing up this concern about model provider flexibility. I completely understand the desire to use local models and custom endpoints.

I want to clarify that we're using the OpenAI Agents SDK by design, not as a limitation. We chose this framework because it gives us a reliable foundation for our agent-based architecture while actually providing support for multiple providers.

The OpenAI Agents SDK has official extensions that let us connect to other model providers through their AI SDK system. You can find the details at https://openai.github.io/openai-agents-js/extensions/ai-sdk/. This means we can already support providers like Anthropic Claude, Google Gemini, Azure OpenAI, Mistral, and even custom OpenAI-compatible APIs - local ones - by installing the right packages and changing configuration.

For local models and self-hosted LLM solutions, I'm also planning to create a completely separate branch that will focus specifically on local deployments and make the app fully free.

So overall there will be two solution - a reliable cloud-based version for users who want that, and a completely free local version for users who prefer self-hosted solutions. Both will coexist so users can choose what works best for them.

First thing we should be doing is implement multiple providers, https://openai.github.io/openai-agents-js/extensions/ai-sdk/. I'll share the update here once its completed.

[trekGriffin]

tried to using aisdk with openrouter models. didn't work for me. got:

Argument of type 'OpenRouterCompletionLanguageModel' is not assignable to parameter of type 'LanguageModelV1'. Types of property 'specificationVersion' are incompatible.

I think I need to follow the your official guidance.