Launching today
polyrouter
One endpoint for every model.
2 followers
One endpoint for every model.
2 followers
Open-source, self-hosted LLM gateway. Point your agents at one URL with one key — explicit-first routing with fallbacks, hard spend limits, and honest, metadata-only cost tracking. Your keys, your box, zero markup.
Hi all,
I got sick of only providing a single model to Openclaw/Hermes Agent/Opencode/other harnesses without any sort of intellegent routing paradigms.
Heartbeats and simple cron jobs shouldn't use the same expensive models as complex coding or research requests. Thus, I developed "polyrouter."
The idea is simple: add tiers, fallbacks, etc, and this app will intellilgently route each request to models as you define, based on complexity - fingerprinting and removing harness system prompts (etc) so they don't get accidentally considered as "complex" - only focusing on the actual pure prompt request itself.
Full disclosure: I did take inspiration from Manifest (https://manifest.build), however they are deprecating one of their main functionalities for this purpose - rule-based routing - so I developed this app instead, while simultaneously improving the feature.
Self-host as you see fit (Docker recommended) and please provide any constructive feedback/feature requests/improvement requests - would love any and all suggestions for improvements!