Orchestrate your inference
Several models. One pipeline. Plain Python.
Chains on OneInfer Cloud let you build compound AI systems: a Python-first framework for complex multi-model pipelines. A small model can classify each input, plain Python decides which path it takes, and a vision or language model answers it.
You write the control flow with the if-else, loops and async/await you already use, and each stage of a chain scales on its own instead of with the whole pipeline. The example chain shows the idea: each input is classified first, then sent to the model that fits it.
An example chain
- Input with an imageBranches to the vision model.
- Text-only inputBranches to the language model.
- Scanned documentAlso goes to the vision model.
An example chain. Input with an image: Vision model. Branches to the vision model. Text-only input: Language model. Branches to the language model. Scanned document: Vision model. Also goes to the vision model.
Illustration of control flow between models.
Designed for complex pipelines
Low overhead
Built on gRPC and zero-copy serialization, so passing work between the stages of a chain adds as little as possible to your multi-model workflows.
Python first
Use familiar Python control flow. If-else, loops, and async/await work exactly as you expect in production.
Native scaling
Each stage in your chain scales independently. The orchestrator automatically handles model placement and traffic routing.
Build on OneInfer Cloud
Create an account to start building, or see everything OneInfer Cloud runs.