OrcaRouter provider support for Agent Zero

Agent Zero stands out because it gives the agent a real Linux desktop inside the Canvas instead of a chat-only loop. Watching it drive real GUI software like Blender or turn a webpage into an instruction surface with the Browser's DOM annotate mode is a workflow that needs more than chat: real tools, live documents, and you able to intervene whenever something looks off.

That autonomy is why model choice matters, and Agent Zero treats providers as a first-class concern: conf/model_providers.yaml keeps a curated catalog, Model Presets bind main, utility, and embedding models to a shared per-provider API key, and calls route through LiteLLM. Because that abstraction already exists, an OpenAI-compatible endpoint fits naturally, and users running long autonomous sessions across projects gain most from more options.

Proposal: OrcaRouter as an optional provider

I propose adding OrcaRouter as an optional model provider. It is a model gateway with a single OpenAI-compatible endpoint: one API key reaches many chat, reasoning, image, and video models from multiple vendors. It would be a new choice alongside Anthropic, DeepSeek, Cerebras, and the rest in conf/model_providers.yaml, not a replacement for or change to any existing provider.

Why it fits Agent Zero's users

Sessions combine a reasoning main model, a lighter utility model, and an embedding model for memory, often running unattended for a long time. The most relevant capabilities are:

• One endpoint to many models, so a Model Preset can move between vendors without per-provider api_base edits.
• Automatic routing and provider failover, so a degraded upstream provider does not stall an unattended session.
• Usage tracking and budgets per API key, useful when running several projects or sharing an instance.

Expected integration point

OrcaRouter exposes an OpenAI-compatible API and uses standard API-key authentication. Given how Agent Zero is structured, the natural integration point is a new entry in conf/model_providers.yaml mapped to litellm_provider: openai, an api_base for the OrcaRouter endpoint, and credentials following the existing one-environment-variable-per-provider pattern (e.g., an ORCAROUTER_API_KEY), with models surfacing through the Model Presets UI. I have not written or tested any code against the repo; this is a proposal for maintainers to weigh in on.

Open-source ecosystem

OrcaRouter is already used by open-source projects, including agent frameworks like goose and CAMEL and the Dify LLMOps platform.

Disclosure

I'm an engineer on the OrcaRouter team. OrcaRouter also runs an optional open-source partner program: 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 is entirely at Agent Zero's discretion; I am happy to follow the project's disclosure or governance requirements, or to leave the topic out if preferred.

You can see existing integrations at https://www.orcarouter.ai/built-with. I would welcome maintainers' thoughts on whether this fits Agent Zero's roadmap, and with approval I would gladly submit an implementation PR following the contribution guide.