Quick answer
LangChain Managed Deep Agents is a public-beta deployment path for the open-source Deep Agents harness. You keep the agent project—model, instructions, tools, skills, middleware, and business logic—in your repository. LangSmith supplies the harness and hosted runtime, including durable execution, persistence, tracing, sandboxes, memory, channels, schedules, and deployment infrastructure where configured.
The current gate is narrow: LangChain documents availability on LangSmith Cloud in the US region only, with a CLI-first workflow while the supported API is finalized. Beta means the feature works end to end, but interfaces, behavior, pricing, deployment options, and SLA coverage can still change.
What the public beta includes
| Layer | Your responsibility | Managed surface |
|---|---|---|
| Agent behavior | Model, instructions, tools, skills, middleware, subagents, domain logic | Deep Agents harness integration |
| Runtime | Test behavior and define failure policy | Durable execution, streaming, persistence, retries, and hosted Agent Server |
| Working environment | Tool policy, filesystem rules, command limits | Optional LangSmith Sandboxes and lifecycle |
| State | Decide thread and memory scope | Thread persistence and optional Context Hub-backed durable memory |
| Access | Choose caller authentication and authorization | LangSmith API-key default or documented end-user identity path |
| Operations | Define evals, alerts, response, and rollback | LangSmith traces, deployment, schedules, and eval packaging |
These capabilities do not remove application responsibility. A hosted runtime cannot decide which tools are safe, which data may be retained, who can act for whom, or what business action should be reversible.
Current setup path
The Python quickstart currently requires an organization with public-beta access, a LangSmith API key, Python and uv, and an API key for the selected model provider. The documented flow is:
uv tool install managed-deepagents
mda init research-assistant
cd research-assistant
uv sync
mda dev .
mda deploy .mda dev starts a local Agent Server and opens the project in LangSmith Studio. mda deploy packages the project and creates a hosted LangSmith deployment. Keep .env out of version control, move production credentials to approved secret storage, and verify which values the deploy step transfers.
The launch announcement also describes TypeScript authoring through the same mda workflow. Confirm the language-specific current documentation, package version, and generated project shape before standardizing a team template.
Beta boundary checklist
Do not approve a production rollout until the team has current evidence for:
- US-region data, legal, residency, latency, and availability requirements.
- Beta change tolerance, version pinning, support path, pricing, quotas, and rollback.
- LangSmith and model-provider credential custody, rotation, revocation, and log redaction.
- Caller identity, private-thread isolation, tool authorization, and downstream reauthorization.
- Memory scope, retention, deletion, provenance, and protection from cross-user disclosure.
- Sandbox scope, network and tool permissions, timeouts, cleanup, and incident isolation.
- Trace and payload logging, access, retention, export, deletion, and sensitive-data handling.
- Evals for tool choice, state changes, artifacts, interruptions, retries, and recovery—not only final text.
Use the production primitives guide to review state and identity. Compare deployment ownership in Managed Deep Agents vs. LangSmith Deployment vs. self-hosting, then record the decision with the readiness checklist.
Frequently asked questions
Is Managed Deep Agents generally available?
No. LangChain documents Managed Deep Agents as a public beta on LangSmith Cloud in the US region only.
Does Managed Deep Agents choose the model for you?
No. You select and configure the model and provide the model-provider credentials, instructions, tools, skills, middleware, and other business logic.
Can teams deploy without the mda CLI?
The current public-beta scope is CLI-first while LangChain finalizes its supported API. Recheck the documentation before depending on another deployment method.
Official sources
- LangChain: Managed Deep Agents public beta
- LangChain: Managed Deep Agents quickstart
- LangChain: release stages
Source check: August 12, 2026. Recheck release stage, regions, access, CLI and API support, packages, prerequisites, pricing, limits, and support terms before deployment.