Azure Container Apps now supports dynamic sessions in public preview. This fast, sandboxed, ephemeral compute is suitable for running untrusted code at scale in hostile multi-tenancy scenarios. Each session has full compute isolation using Hyper-V.
Two modes are supported for dynamic sessions: code interpreter and custom container.
- Code interpreter sessions:
Dynamic sessions enables fast and simple access to sandboxed code interpreters for executing untrusted code. It allows you to run Python code and is preinstalled with many popular Python packages. More languages and runtimes are planned in the future.
To build advanced AI agents or copilots, large language models (LLMs) are often paired with a code interpreter. A code interpreter enables an agent or copilot application to extend an LLM’s abilities to perform complex tasks such as solving mathematical and reasoning problems, analyzing data, and generating graphics and charts. Dynamic sessions’ code interpreter provides sandboxes to securely execute LLM-generated code in production.
You can build your own copilots by adding code interpreter sessions to popular frameworks like LangChain, LlamaIndex, and Semantic Kernel in a just few lines of code.
- Custom container sessions:
Dynamic sessions also lets you run your own custom container. You can use a custom container to create a code interpreter for your specific needs, such as preinstalling dependencies or supporting a different language.
You can also build a custom container to run any application that requires secure sandboxed environments. Some example scenarios include:
- A web-based code editing sandbox
- A SaaS application that can be extended by end users with custom code snippets
- A hosted Jupyter Notebook service
