Skip to main content
This page helps you understand how LangChain Deep Agents compare to the Claude Agent SDK and the Codex SDK.

Overview

AspectLangChain Deep AgentsClaude Agent SDKCodex SDK
Primary use caseInteractive coding agent in terminal or IDE; Build production agents programmatically with SDKBuild production agents programmaticallyInteractive coding agent (CLI, desktop, IDE, cloud); Build production agents programmatically with SDK
Model supportModel-agnostic (Anthropic, OpenAI, and 100s others)Claude models (Anthropic, Azure, Vertex AI, AWS Bedrock)OpenAI models (GPT-5.3-Codex and variants)
LicenseMITMIT (underlying Claude Code is proprietary)Apache-2.0
ArchitecturePython SDK, TypeScript SDK, and CLIPython SDK, TypeScript SDKTypeScript SDK, CLI, desktop app, IDE extension, cloud interface
FrontendIntegration with ReactServer-side onlyServer-side only

Key differences

LangChain Deep Agents:
  • Model flexibility: Use any provider (Anthropic, OpenAI, or 100+ others) and swap models without changing your agent logic
  • Long-term memory: Persist context across sessions and threads with the Memory Store
  • Sandbox-as-tool pattern: Run individual operations in isolated sandboxes while the agent runs outside, or run the full agent inside a sandbox
  • Virtual filesystem: Use pluggable backends (in-memory, disk, durable stores, sandboxes) for context and checkpoint management
  • Production deployment: Deploy via LangGraph with built-in hosting options
  • Observability: Use LangSmith for native tracing and debugging
Claude Agent SDK:
  • Standardize on Claude: First-class support for Claude models across Anthropic, Azure, Vertex AI, and AWS Bedrock
  • Custom hosting: Build your own HTTP/WebSocket layer and run the SDK in containers
Codex SDK:
  • Standardize on OpenAI: GPT-5.3-Codex and OpenAI-specific tooling
  • OS-level sandbox modes: Use built-in read-only, workspace-write, or danger-full-access modes
  • MCP server mode: Expose your agent as an MCP server with codex mcp-server
  • Observability: Use OpenAI Traces

Feature comparison

FeatureDeep AgentsClaude Agent SDKCodex SDK
Core tools
File Read/Write/Editls, read_file, write_file, edit_fileRead, write, edit✅ Built-in
Shell Executionexecutebashexec
Glob/Grepglob, grepglob, grep✅ Built-in
Web Search✅ support for third-party and provider-nativeWebSearch, WebFetchweb_search
Planning/Todoswrite_todosTodo listsPlan before changes
SubagentsSubagentsSubagentsMulti-agent workflows (experimental) via MCP + Agents SDK
MCP ClientMCPMCP client
Human-in-the-LoopApprove/edit/rejectPermission modesApproval modes (Full Access, Read-only, Auto)
Skills SystemSkillsSkillsSkills
Additional instructionsMemoryCLAUDE.md filesAGENTS.md
Long-term memoryMemory Store
StreamingStreamingStreamingStreaming
Sandbox Integrations
Agent runs in sandboxCan run in sandboxCan run in sandboxCan run in sandbox
Agent runs operations in sandboxesUses sandboxes as tools
OS-level sandbox modesread-only, workspace-write, and danger-full-access
Architecture
Composable MiddlewareHooks
Virtual FilesystemsVirtual filesystem with pluggable backends
Agent Server
Production hostingLangGraph deployment, self-hosted, or standaloneBuild your own HTTP/WebSocket layer; run SDK in containersCodex App server
ACP server (IDE integration)ACP server✅ (third-party)✅ (third-party)
MCP server modeMCP endpoint through Agent Servercodex mcp-server
A2A endpointA2A endpoint in Agent Server
State management
Session ResumeSession managementResume threads; SDK resumeThread()
File CheckpointingBackends and checkpointsFile checkpointingGit checkpoints
Time Travel (state branching)
Observability
Native TracingLangSmithOpenAI Traces

Notice a mistake?

We drafted this comparison on February 18th, 2026. If you notice mistakes or changes in products, please file an issue.
Connect these docs to Claude, VSCode, and more via MCP for real-time answers.