LangChain DeepAgents Deploy: Config-Driven Cloud Deployment for Agent Harnesses
LangChain has shipped config-driven cloud deployment for its DeepAgents harness via a deepagents.toml file with four sections. The [agent] section names the deployment in LangSmith and selects from models across OpenAI, Anthropic, Google, OpenRouter, Fireworks, Baseten, and Nvidia. The [sandbox] section connects to a code execution environment (Modal, Daytona, or Runloop). The [auth] section enables multi-tenant deployments via Clerk and Supabase with RBAC for operator access. The [frontend] section delivers a streaming UI with thread history and pre-wired endpoints. Madrigal Pharma is already using DeepAgents for a multi-agent deep research system.
Why It Matters
A single config file abstracting model selection, sandboxing, auth, and UI for cloud agent deployments reduces the infrastructure engineering burden for teams building production agents—directly competing with AWS AgentCore and custom build-it-yourself stacks.