Weave Router
A self-hosted proxy that automatically routes each AI request to the best-performing, cheapest model across Anthropic, OpenAI, Gemini, and open-source providers.
🔗 Visit Weave Router
Description
If you use several AI models across different providers — Claude for one task, GPT for another, an open model for a third — you either hardcode which model handles what, or you eat the cost of always calling the most expensive one "just in case." Weave Router sits in front of your AI calls as a drop-in proxy and picks the best model for each individual request automatically, aiming to cut costs by 40-70% without you having to manually manage the routing logic.
Weave Router uses a cluster-scoring approach (based on "Avengers-Pro" research) to make per-request routing decisions across Anthropic Messages, OpenAI Chat Completions, Gemini, and open-source models via OpenRouter, with claimed sub-50ms added latency and streaming support. API keys are encrypted and stored locally per customer (bring-your-own-key), and it ships with built-in OpenTelemetry tracing and a local dashboard (localhost:8080/ui) for observability. It's written in Go, backed by Postgres, distributed as a self-hosted Docker deployment or an npm package (@weave-os/router), and licensed under the Elastic License v2 — free to self-host, but restricted from being resold as a competing hosted service. One functional limitation: streaming responses can't include the final cost figure in HTTP headers, since the cost isn't known until the stream completes.
💬 Our review
The short version: Weave Router is a genuinely useful piece of AI infrastructure if you're already juggling multiple model providers and care about cost — its 3,900+ GitHub stars suggest real adoption — but it adds a Postgres-backed service to your stack, which is real operational overhead for a solo developer or small team.
Against just picking one provider and living with its pricing, Weave Router's whole value proposition is arbitrage across providers, which only pays off if your usage volume is high enough that a 40-70% cost cut is meaningful in absolute terms. Against a comparable hosted router like OpenRouter, Weave Router's key difference is that it's self-hosted and keeps your provider API keys local and encrypted rather than passing everything through a third party's infrastructure — a real advantage if you have data-residency or compliance concerns, at the cost of running and maintaining the Postgres-backed service yourself. It's free to self-host (Elastic License v2), so the real cost is the ops overhead, not money. Worth it for teams with meaningful multi-provider AI spend who want to keep routing in-house; overkill if you only call one model provider.
💰 Pricing
📊 Global score
🤖 AI-enriched data
Auto-hébergement gratuit, licence Elastic License v2 (revente en service concurrent interdite) ; option service hébergé géré également proposée
Pros
Routage automatique par requête vers le modèle le moins cher/plus performant, réduction de coût annoncée de 40 à 70%
Auto-hébergeable, clés API chiffrées et stockées localement (bring-your-own-key)
Observabilité intégrée (traces OpenTelemetry, dashboard local)
Plus de 3900 étoiles GitHub, adoption réelle
Cons
Nécessite une infrastructure Postgres à maintenir soi-même
Licence Elastic v2 : gratuit à auto-héberger, mais pas open-source au sens strict
Le coût final n'est pas inclus dans les en-têtes HTTP en streaming (connu qu'à la fin du flux)
N'a de sens que si le volume d'appels IA multi-fournisseurs est déjà significatif