deep-agents-core

Build multi-step LangChain agents by deploying a reusable agent harness with built-in TodoList middleware for task planning and Store backends for persistent memory. This approach lets you configure complex orchestration instead of implementing it from scratch.

Manage large context for deep agents by configuring a harness with filesystem and store backends. This enables efficient file tool loading and persistent memory across threads for long-running investigations.

Yes, you can add human-in-the-loop approvals to an agent workflow by leveraging checkpointer-based interrupt support. This allows your application to pause execution and request human approval before sensitive write operations occur.

Subagent delegation works by allowing a primary deep agent to spawn specialized subagents for distinct functions like data collection and code synthesis. This modular delegation is configured through the agent harness for production deployment.

You need a store backend like an InMemoryStore to enable long-term memory across threads. The agent harness supports configurable filesystem and store backends to persist findings across different sessions.

Use a deep agent harness instead of custom orchestration when your workflows require task planning, subagent delegation, or human-in-the-loop approvals. It provides middleware-driven planning and checkpointer integration to avoid implementing complex orchestration from scratch.