OrcaRouter provider support for Anton

Anton is a doing agent rather than a coding agent: you describe the outcome you want — a cleaned inbox, a live portfolio dashboard pulled from current prices, a scheduled report — and Anton figures out the steps, writing whatever code is needed along the way. Its multi-layer memory means lessons learned on one task carry into your next workflow, which is what makes it genuinely self-improving.

What also stands out is how model-agnostic Anton is: its setup already supports Anthropic and OpenAI BYOK, MindsHub's Model Router, and generic OpenAI-compatible endpoints such as Together, Groq, Ollama, and vLLM, and it can split planning from coding across different models. For anyone running Anton standalone with their own key, each extra OpenAI-compatible option is a real choice — more model availability, more resilience, and more control over cost.

I'm an engineer on the OrcaRouter team. This is a proposal — no code has been written or tested yet. OrcaRouter would be an optional, additional provider choice; it would not replace or alter any existing provider, and the MindsHub Model Router path you already expose would stay untouched.

OrcaRouter exposes an OpenAI-compatible API and uses standard API-key authentication, so it should fit the same provider/config abstraction Anton already uses for generic OpenAI-compatible endpoints (a base URL plus an API key, via the /setup Models wizard or .anton/.env). The expected integration point is exactly there — alongside Together, Groq, Ollama, vLLM, and similar.

Which users would this help? A few capabilities map naturally onto Anton's workflow:

• Multiple chat, reasoning, image, and video models through one endpoint, so your planning/coding provider split stays stable while the models underneath can change.
• Automatic model routing and provider failover, which matters for long-running or scheduled workflows where a single provider outage shouldn't kill a task.
• Prompt caching, which cuts cost on the repetitive multi-turn loops agentic work generates.
• Usage tracking and budgets, useful when several Anton sessions share one key.

OrcaRouter is already used with a number of open-source projects — Dify, RAGFlow, goose, and promptfoo among them — and we maintain a gallery of those integrations at https://www.orcarouter.ai/built-with.

One transparent disclosure: OrcaRouter runs an optional open-source partner program through which approved OSS projects can receive a 5% revenue share from OrcaRouter usage attributed to their integration. Participation is not a prerequisite for the integration, and I'm glad to follow whatever disclosure or governance rules Anton prefers — or to leave the program out entirely.

I'd welcome the maintainers' thoughts on whether an OrcaRouter option would be useful for Anton's users and where it would best slot into provider setup. If it seems like a good fit, I'd be glad to open an implementation PR.

Start by locating the existing provider/config abstraction used by Together, Groq, Ollama, and vLLM, then trace how the /setup Models wizard and .anton/.env represent generic OpenAI-compatible endpoints. Done should mean OrcaRouter is an optional provider alongside the existing choices without changing the MindsHub Model Router path, with the integration behavior tested.