This is an early release preview. You may encounter bugs.
OpenShell DeepAgent logo
Unclaimed

Harness coding security

OpenShell DeepAgent

General-purpose coding agent pairing LangChain Deep Agents with an NVIDIA OpenShell sandbox and Nemotron models

Built by LangChain

C 63/100 GitHub score ? This grade is derived from GitHub signals, not user votes. Open for the full breakdown.
No votes yet

01 / About

What OpenShell DeepAgent is.

OpenShell Deep Agent is a general-purpose coding agent that runs inside an NVIDIA OpenShell sandbox, orchestrated by LangChain's Deep Agents library and powered by NVIDIA Nemotron. The agent writes and executes code in an isolated, policy-governed Linux environment on your own machine.

OpenShell is an on-prem sandbox for agent code execution that enforces policies over filesystem access, network access, and process permissions, so something external to the agent, rather than prompt instructions alone, prevents misuse. Deep Agents supplies the harness: the standard tool-calling loop plus built-in task planning, a file system for context management, subagent spawning, and long-term memory, running on the LangGraph runtime.

A LangGraph dev server hosts the agent with execute, writefile, readfile, edit_file, glob, grep, and ls tools. Code execution goes through an OpenShellBackend that talks to the sandbox container over gRPC, while memory (/memory/AGENTS.md) and skills (/skills/*.md) live on local disk through a FilesystemBackend so they persist across restarts. The backend is a drop-in replacement for the Modal backend, so memory, skills, and subagents are unchanged.

Features

  • Sandboxed execution: scripts written to /sandbox/ run inside the OpenShell container via SandboxSession.exec()
  • Policy enforcement: a request to exfiltrate data to an outside host is blocked by the sandbox network policy regardless of instructions
  • Hot-reloaded policies: export the YAML policy with openshell policy get, edit it, and push it back with openshell policy set without restarting
  • Model: NVIDIA Nemotron Super 3 via NVIDIA NIM, configured with an NVIDIAAPIKEY
  • Persistent memory and skills: AGENTS.md and skill files on local disk that can be committed to git
  • Studio UI: the LangGraph dev server exposes an API on 127.0.0.1:2024 and a LangSmith Studio link, with optional LangSmith tracing
  • Gateway on Docker: the OpenShell gateway runs k3s inside Docker Desktop and manages named, reusable sandboxes

02 / Discussion CREDIBILITY-GATED

Discussion

Reading is open to everyone. Posting and voting need a verified identity or a GitHub grade of B or higher.

  • No discussions yet.

04 / Build

Build with OpenShell DeepAgent.

Browse the catalogue for frameworks, tools, and harnesses, each scored on real GitHub credibility.

Get OpenShell DeepAgent →

Browse the catalogue