LLM Gateway Router for Agents

A Claude Code compatible gateway that routes subagent calls to different LLMs (on-device, cheaper open models, or premium models) to optimize cost and performance. Targets developers building with Claude agents who want granular control over model routing.

LLM routing is getting real traction as agentic workloads explode and token costs compound fast across multi-step reasoning chains — developers building with Claude Code are already hitting this problem today, not theoretically. LiteLLM is the closest substitute and already handles basic routing, so the defensible angle here is tight Claude agent compatibility and subagent-level granularity that general-purpose proxies don't prioritize. The $2k–10k/mo band is plausible for a dev-tools infrastructure play if it lands even a handful of teams running production agents at scale, since willingness to pay tracks directly with token spend saved. The biggest risk is that Anthropic ships native routing controls into Claude Code itself, which would kill the core value prop overnight with no warning.