Skip to main content
Deep Agents is LangChain’s harness for long-running agents. It takes a pluggable backend for its file and shell tools. Mirage ships LangchainWorkspace, a backend over a Workspace, so every file the agent touches and every command it runs stays in the workspace’s mounts.

Install

Add the LangChain integration package for the chat model in use, such as langchain-anthropic or langchain-openai.

Usage

Tools

The agent keeps Deep Agents’ own file tools, run in Mirage: ls, read_file, write_file, edit_file, delete, glob and grep. LangchainWorkspace is a sandbox backend, so the agent also gets execute, which runs a command line in Mirage’s shell. Every call acts as the session (session_id, default the workspace’s own), so its profile and the mount modes judge it. read_file passes images, PDFs, audio, video and PowerPoint files to the model as multimodal content, when the model supports that input. Text files keep line-based pagination.

Exports

Examples