Caution
There is no current release, I am still building the MVP.
mockan is a mock API server that helps teams work with Machine Learning/AI workflows. Teams can do experimentation without incurring into high costs and without waiting for the models to be ready, or simulate incidents.
It simulates a working API (input, ouput, queues, delay and saturation/errors) of multiple cloud AI services (OpenAI, Anthropic, AWS Bedrock) and self-served (NVIDIA Triton Inference Server, Torchserve, TensorFlow Serving). There are many included mock models and you can create new ones easily.
- Any team can create multiple scenarios and learn how each sub-system will react to outages, saturation or broken deployments without bugging the ML team and without asking for resources.
- Future-proof your system. Check how eventual success will impact your job activities.
- You don't have to wait for an API key or an ML engineer to create the service.
- Prepare for the integration tests
Following the Diátaxis framework developed by Daniele Procida, the documentation is split into 4 categories:
You can't (yet).