Show HN: Claude Code 2.0 router – preference-aligned routing to multiple LLMs (opens in new tab)

Hi HN — we’ve been working on a small project that routes between different LLMs based on user-defined preferences rather than just benchmarks or latency. The earlier version (Arch-Router [1]) was a 1.5B model that matched queries to domains (e.g., travel, creative writing) or action types (e.g., editing).

We’ve now applied the same idea to coding workflows, packaged into a CLI agent via Arch Gateway [2]. The tool gives:

Unified model access — use Claude Code, Grok, Mistral, Gemini, DeepSeek, GPT, or local Ollama models from the same interface.

Task-specific routing — assign different models for tasks like code generation, reviews, debugging, or system design.

Instead of relying on public benchmarks (MMLU, MT-Bench, etc.), routing is guided by user-defined criteria. This helps capture subjective quality, task relevance, and individual developer preferences that generic metrics often miss.

Links:

[1] https://huggingface.co/katanemo/Arch-Router-1.5B [2] https://github.com/katanemo/archgw