LangChain Deep Agents CLI lands as terminal agent

This feels like LangChain's open-source answer to Claude Code, but with a more explicit bet on model neutrality and inspectable, file-backed state.

• –Model switching is first-class, with OpenAI, Anthropic, and Google support out of the box and more providers available as extras.
• –Memory lives in files under `~/.deepagents/...`, so project conventions are auditable instead of trapped in hidden state.
• –Skills, MCP tools, and custom subagents turn the CLI into a composable harness, not just a prompt wrapper.
• –Approval controls and sandboxes matter because this agent can edit files and run shell commands.
• –The tradeoff is surface area: a more powerful terminal agent also means more setup and policy decisions than a closed one-off tool.