LangChain releases deep agents open source tool for context management

Summary

LangChain introduced Deep Agents, an open-source agent tool designed to combat context rot through offloading, summarization, skills, subagents, and long-term memory. The framework stores agent memory and offloaded data as plain files in user-controlled storage, allowing developers to switch AI models or providers without losing accumulated learnings.

What changed

• Offloads large tool results and user messages to the filesystem while displaying short previews to the model.
• Summarizes older messages once token usage reaches 85 percent of the context window.
• Loads specialized skills and task instructions progressively only when deemed relevant.
• Isolates noisy background work in separate subagent context windows and persists long-term memory in markdown files.

Availability

“uv add deepagents”

Source