What problem does it solve?
Deep Agents Core removes the friction of assembling multi-step AI assistants by providing an opinionated harness and configuration surface so you configure capabilities instead of implementing them from scratch. It centralizes task planning, context management, delegation, persistent memory, skills loading, and human-in-the-loop approval into a reusable agent architecture.
Core Features & Use Cases
- Task Planning: Built-in TodoList middleware for breaking complex requests into actionable steps.
- Context Management: Filesystem or Store backends to manage large, file-backed context and enable on-demand skill loading.
- Task Delegation: SubAgent middleware to spawn specialized agents for research, coding, or automation subtasks.
- Long-term Memory & Interrupts: Pluggable Store and checkpointer support for persistent threads and safe interrupt handling.
- Human-in-the-loop & Safety: Approval workflows and configurable interrupt rules for sensitive operations.
- Use Case: Build a persistent research assistant that manages long-running investigations, spawns subagents for data collection and code execution, and stores results across sessions.
Quick Start
Create a deep agent configured with your model, tools, backend (filesystem or store), skills directory, and checkpointer to run a persistent multi-step assistant.
