This application demonstrates how to use the deepagents package to build a coding assistant that can run as a LangSmith Deployment.
deepagents gives you a powerful harness for creating agents that can handle complex, multi-step tasks, such as
software engineering projects. The LangSmith Deployment platform gives you a production-ready server
and managed infrastructure to run such a complicated agent scalably and reliably as you increase the scope
of its work. This repo shows how you can combine these two LangChain offerings to design and deploy a software
engineering agent that can do useful work in the unique context of your organization.
You can use the agent via a chat interface - it is prompted to return its final code output in its response
The agent learns about the user's preferences for code and saves them to LangSmith Deployment's managed long-term memory store. The store keeps track of separate preferences per Assistant
The skills stored in the codebase are uploaded to the sandbox so the agent can use them and run any associated code. There are some example skills in this repo.
This agent can write and test code within a Daytona sandbox. For simplicity, the agent will return the final code it has written as a message to the end user, but it can iterate on and test the code before deciding it is done.
You can modify this code to instead use Runloop or Modal, or implement your own sandbox.
You will need the following environment variables:
You can run this locally using LangSmith Studio or deploy this code to a LangSmith Deployment