Documentation reviewed
Managed Deep Agents
Checked
Recheck due
Managed Deep Agents runs code-defined agents on LangSmith Cloud. It supplies the Deep Agents harness and managed runtime for threads, tool use, sandbox work and persistent context. The October 29 v0.10 release adds browser use and asynchronous subagents.
- Availability
- Beta. Public beta since the publisher's August 7 update. v0.10 is an update to that beta, not a new GA product. Browser sessions and parallel work still require scoped permissions and approval design.
- How to use it
- API; Library / CLI; Web app. Author a Python or TypeScript project with the mda CLI and inspect it in LangSmith Studio. Hosted deployment choices and tool availability vary by SaaS region.
- Processing and data
- Vendor cloud. Agent code, configured secrets, context and execution state reach managed services; model and tool providers add their own destinations. Browser handoff and thread/user scopes are documented behavior, not an independent confidentiality guarantee.
- Costs and limits
- Deployment plan and usage; model and third-party tool charges are separate
- Outputs
- Hosted agent runs, streamed responses, tool actions and sandbox files. Async subagents share the agent sandbox; do not assume isolated files between helpers.
Public beta, not GA. Requires a LangSmith workspace, model-provider credentials and deployment entitlement. The pricing table places Deployment on Plus/Enterprise; confirm beta entitlement and usage before setup. Managed tools are GCP US only, and Serverless needs a supported GCP environment. No deployment, billing or isolation test was performed.