Added
Updated
Conifer
Conifer routes AI queries across models to cut token spend sharply.
Founded
2026Conifer routes AI queries across models to cut token spend sharply.
Conifer sells a least-cost routing layer for large language model inference. Its gateway exposes one OpenAI-compatible and Anthropic-compatible endpoint at api.conifer.build that fronts more than 200 models, and a router decides for each query whether it is served on the customer's own hardware or on a cloud provider.
The company also ships a console for keys, spending, and billing, and an open-source SDK and command-line runtime under its ConiferKit GitHub organization. Conifer was founded in 2026, is part of Y Combinator's Summer 2026 batch, and is based in San Francisco.
Investment in large language model applications has shifted spending from model access toward inference cost control, and routing layers sit at that seam. Conifer competes in an active field of AI gateways and model routers where differentiation comes from how far down the cost curve a router can push a workload without changing the model a developer asked for.
Local-first routing is the company's stated wedge: serving queries on hardware the customer already owns removes the per-token fee entirely. Adoption depends on developers accepting a routing layer that decides between their own machine and a cloud provider.
Conifer's routing spans local hardware and cloud providers in one decision path, so a query can be served on the customer's own machine at no API fee before it falls back to an efficient cloud model and then to a frontier model. Customers keep their existing harness and tooling because the gateway speaks the OpenAI and Anthropic wire formats.
The company prices as prepaid pay-as-you-go at catalog rates and charges no fee when a customer brings its own provider keys. Named models are served as named, with the router choosing the venue rather than substituting a different model.