LowRouter
LLM router hosted in France, with a carbon figure on every call
LowRouter is an OpenAI-compatible router that sends requests to model providers such as Mistral, Scaleway, OpenAI, Anthropic and Groq with one API key and one EUR invoice. Tokens are billed at the upstream provider's price with no per-token markup; the fee is taken when credits are bought. In auto mode the router prefers EU-sovereign routes, then the lowest grid carbon intensity, then price. Each response carries the provider, region and an estimated gCO2eq. Operated by Carbonifer SAS in Nantes. In beta, with prompts not retained.
Pricing: Pay-as-you-go
Choosing an LLM router right now?
One click, anonymous. Helps us know our visitors.
Thanks, that helps.
Choosing an LLM router right now?
One click, anonymous. Helps us know our visitors.
Thanks, that helps.
What is LowRouter?
LowRouter is an OpenAI-compatible router operated by Carbonifer SAS in Nantes, France. One API key reaches models from providers such as Mistral, Scaleway, OpenAI, Anthropic and Groq, and billing is one EUR invoice. It does not support bringing your own provider keys. The service is in beta.
How routing works
Every request names a model in one of three ways, per the routing docs:
- Explicit model: a full model ID. The router resolves it to the provider and region encoded in the ID.
- Auto-routed model:
auto/<creator>/<model>. You pick the model; the router picks which provider and region serve it. The weights and output are the same canonical model. - Alias:
alias/<name>, one of your account's stored routes.
The router chooses the route, never the model. An earlier lowrouter/auto that also chose the model was removed, because ranking on carbon alone would always have picked the smallest model.
In auto mode the default order is:
- EU-sovereign routes first. A route qualifies only when the provider is an EU-sovereign company and the serving region is in the EU/EEA.
- Lowest grid carbon intensity. A region with no grid data ranks below every region with data.
- Lowest price.
- Round-robin across what is still tied.
Each API key can have its own routing policy with a different order.
Sovereignty is a preference unless you scope it
In auto mode, sovereignty is a preference. If a model has no sovereign route, the request falls through to the lowest-carbon, then cheapest route anywhere, which can be a non-EU provider. For a hard rule, append a scope such as auto/mistralai/mistral-large-2512/eu-sovereign, use an explicit route or set a key policy. The request then fails instead of leaving the scope. Other scopes include eu, europe, cloud-act-free, us and country codes. Every response carries lowrouter_metadata with the provider, region, whether the route was EU-sovereign and the step that decided.
The carbon figure is an estimate
Each response reports energy in Wh and carbon in gCO2e, from the published methodology. Energy comes from the EcoLogits v0.4 inference model. Grid intensity comes from Ember country averages, replaced by regional grid-operator data where an operator publishes it. Two limits the docs state: only generated tokens are counted, so requests that are mostly prompt are under-counted, and no validation study against measured energy has been run. Each estimate carries an ordinal confidence label (accurate, medium or gross), and LowRouter does not publish a percentage error.
Pricing
Tokens are billed at the upstream provider's price with no per-token markup. The fee is taken when credits are bought: 5.5% + EUR 0.30 for a business with an EU VAT number, 6.5% standard and 8.5% outside the EU/EEA, with a EUR 5 minimum top-up. Checked against the pricing page on 2 November 2026.
LowRouter Alternatives
Explore 173 products in the Inference APIs category. View all LowRouter alternatives.
EUrouter
European AI gateway that routes to 100+ models with EU data residency
TrustRoute
OpenAI-compatible LLM gateway hosted in Switzerland, with per-workload rules on which models data may reach
Runcrate
On-demand GPUs and inference for AI teams, and for the agents they run
inference shell
One API key for any AI model, with skills, agents and a CLI
Work on LowRouter? Feature it at the top of Inference APIs.
Is your product missing?