Runtime for Deep Agents

Deep Agents support is coming soon. When it lands, each orchestration will run in its own cloud environment, with durable state across dozens of steps, scoped network access to external services, isolated sub-agents, and a clean recovery model when long workflows fail late. Today, OpenClaw and Hermes already run on this same per-agent environment model.

Coming Soon

What multi-step orchestration needs from a runtime

Deep Agents workflows run longer and carry more state than single-turn tasks. Each piece below has to land before Spinup can call it supported.

Nothing lost between steps

Multi-step workflows accumulate findings, generate files, and build state over many turns. The work is to define a clear state and recovery contract.

Scoped access to external data

Workflows that reach external services through custom tools or MCP need controlled connectivity. That network behavior needs an explicit shape.

Sub-agents in their own environments

Orchestration chains may spawn sub-agents or branch into parallel paths. Each child workload needs its own isolation and lifecycle rules.

Resume, don't restart

A late failure should not mean starting over. Recovery semantics need to be in place before Spinup claims Deep Agents support.

Orchestration in the Cloud

Why ephemeral sandboxes break multi-step workflows

Single-turn coding agents can work in ephemeral environments. The agent receives a prompt, writes code, and returns a result. If the environment resets between turns, the cost is low.
Deep Agents workflows are different. An agent might plan a complex task, spawn sub-agents, generate intermediate files, and synthesize results across dozens of steps. Each step builds on the last. An environment reset mid-chain means starting over.
You can't run Deep Agents on Spinup yet. OpenClaw and Hermes are the supported harnesses today, on the same isolated microVM and persistent state model. Multi-step orchestration builds on that foundation.

FAQ

Common questions about Deep Agents on Spinup

Why does Deep Agents need persistent environments more than single-turn agents?+

Multi-step workflows accumulate state, generate files, and build context across dozens of turns. An ephemeral sandbox throws all of that away. Persistent environments are why Deep Agents needs to wait for a runtime that holds state between steps. Spinup ships that model with OpenClaw and Hermes today, and Deep Agents lands on top.

How does Deep Agents relate to LangChain?+

Deep Agents is part of the LangChain ecosystem, with implementations in both Python and JavaScript. Dependency, configuration, and run behavior all need explicit work before it can run on Spinup. Today, OpenClaw and Hermes are the supported harnesses.

How does Spinup handle network access for multi-step workflows?+

Not yet. Workflows that reach external services through custom tools or MCP need controlled connectivity, and that contract is part of the Deep Agents work that comes after launch.

Can I use Deep Agents alongside other harnesses?+

Not yet. Today the supported swap is between OpenClaw and Hermes. Deep Agents joins the same portable runtime model when it lands.

Related

The runtime should stay bigger than any one harness

Early access

Deep Agents is on the way. Try OpenClaw or Hermes on the same runtime today.

Request access if this is the runtime shape your team has been missing.

We reply with a booking link and review the fit before inviting your workspace.