Deep Agents CLI

Deep Agents is a framework for building intelligent agents powered by LLMs that can tackle complex, multi-step tasks with built-in capabilities for planning, context management, and task delegation.

The framework provides a complete agent SDK with everything needed to create agents that can handle sophisticated workflows. Unlike basic agent frameworks, Deep Agents includes opinionated defaults and specialized tools designed to make agents more capable and reliable.

Key capabilities include:

• Task planning and decomposition - Agents break down complex problems into discrete steps, track progress, and adapt as new information emerges
• File system management - Offload large context to in-memory or persistent storage, preventing context window overflow and enabling work with variable-length results
• Subagent spawning - Delegate specialized work to isolated subagents while keeping the main agent's context clean
• Shell execution - Run commands for tests, builds, and system operations with optional sandboxing for security
• Long-term memory - Persist information across conversations and threads for continuous learning
• Filesystem permissions - Control which files and directories agents can access with declarative rules
• Human-in-the-loop workflows - Require approval for sensitive operations before execution
• Model flexibility - Works with any LLM that supports tool calling, from frontier models to open-source alternatives

The framework includes pluggable filesystem backends you can swap to match your needs—choose from in-memory state, local disk, LangGraph store for cross-thread persistence, or isolated code execution environments. Deep Agents also ships with smart system prompts that teach models how to use tools effectively, including planning before acting and verifying work.