~/wiki
Récent+ Ajouter

Deep Agents

Mis à jour le 2026-04-14Confiance : high
agent-harnessopen-sourcelangchainmodel-agnosticagent-memoryself-hosting

Open-source agent harness developed by langchain designed to give developers full control over their agent-memory and prevent vendor lock-in. Represents LangChain's answer to proprietary agent platforms.

Core Philosophy

Built on the principle that memory creates competitive moats and lock-in, so developers should own their agent memory rather than yielding control to model providers. Addresses the problem that "if you don't own your harness, you don't own your memory."

Key Features

Open Source: Fully transparent codebase that developers can inspect, modify, and extend.

Model Agnostic: Not tied to any specific model provider, allowing easy switching between OpenAI, Anthropic, and other LLM providers.

Open Standards: Uses standardized formats including:

  • agents.md for agent configuration
  • agentskills.io for skill definitions
  • Standard tool calling interfaces

Memory Store Flexibility: Pluggable memory backends including:

  • MongoDB
  • Postgres
  • Redis
  • Other database systems

Deployment Options:

  • LangSmith Deployment (self-hostable on any cloud)
  • Bring-your-own-database for memory storage
  • Standard web hosting framework compatibility
  • Full control over infrastructure

Contrast with Proprietary Solutions

Unlike closed systems such as:

  • Claude Managed Agents: Everything behind Anthropic's API
  • Claude Agent SDK: Uses Claude Code under the hood (not open source)
  • OpenAI Responses API: Stateful but locked to OpenAI ecosystem

Deep Agents ensures:

  • Complete visibility into how memory works
  • Ability to migrate between providers
  • No encrypted or proprietary memory formats
  • Full data ownership

Technical Architecture

Built on top of langchain and LangGraph, extending their capabilities with:

  • Persistent memory management
  • Context engineering and compaction
  • Tool orchestration
  • State management across sessions

Strategic Positioning

Part of LangChain's broader strategy to prevent the AI ecosystem from being dominated by closed platforms. Addresses the concern that model providers will increasingly move agent capabilities behind proprietary APIs to create lock-in through memory control.

See also