OrcaRouter as an optional model provider for RAI

Is your feature request related to a problem? Please describe.

RAI's vendor-agnostic design makes it a natural home for embodied-AI work on ROS 2. Your config.toml spells that out: a [vendor] selector with [openai], [ollama], [aws] and [google] blocks, each defining simple_model, complex_model and embeddings_model roles, and the docs note that even Ollama can be reached through an OpenAI-compatible endpoint. The orchard-tractor and natural-language Franka Panda demos show the payoff: a robot that reasons and acts through one coherent agent stack.

Robot teams in shared labs or on field fleets today choose between a single hosted provider and a self-hosted model server, with little in between. A hosted, OpenAI-compatible gateway would give those users a middle path, so I would like to propose adding OrcaRouter as an optional provider.

Describe the solution you'd like

I propose adding OrcaRouter as an optional model provider. It would not replace or change any existing provider: OpenAI, Bedrock, Google and Ollama would keep working as today. OrcaRouter exposes an OpenAI-compatible API and uses standard API-key authentication. Since RAI already supports OpenAI-compatible endpoints (the openai block can point at a local Ollama server via base_url) and builds on LangChain chat models, a natural integration point is exposing OrcaRouter as another selectable vendor in config.toml and the RAI Configurator's Model Selection, reusing that OpenAI-compatible path. This is a proposal only; nothing is implemented or tested.

For RAI users, three capabilities stand out:

• Many chat, reasoning and image models through one endpoint, fitting RAI's simple/complex/embeddings model roles that already assume model variety.
• Automatic model routing and provider failover, so a field robot keeps working if one upstream provider degrades, consistent with RAI's vendor-agnostic ethos.
• Usage tracking and budgets, valuable for labs and companies running shared robot infrastructure.

OrcaRouter is already part of the open-source ecosystem, with integration examples including RAGFlow, Dify, DB-GPT and CAMEL, so the pattern is familiar to agent-framework maintainers.

Describe alternatives you've considered

RAI users can already pick OpenAI, Bedrock, Google or self-hosted Ollama. What is missing is a hosted gateway spanning many providers and models behind one API key, which OrcaRouter adds; building equivalent routing, failover and usage tooling in-house would be a lasting maintenance burden.

Additional context

I'm an engineer on the OrcaRouter team. OrcaRouter also runs an optional open-source partner program in which approved OSS projects can receive a 5% revenue share from OrcaRouter usage attributed to their integration. Participating is not a prerequisite for the integration, and I would follow RAI's disclosure and governance requirements. Existing integrations are listed at https://www.orcarouter.ai/built-with. I'd welcome the maintainers' views on whether this fits RAI's roadmap; if so, I'm happy to submit an implementation PR per the contribution guide.

Start with config.toml, the contribution guide, and the RAI Configurator's Model Selection implementation, then trace how existing OpenAI-compatible providers are configured. Done means OrcaRouter is an optional selectable provider with the stated model roles, while existing providers remain unchanged and the integration is covered by tests.