Deep Agents

Headlines · 7

1. 14h
   Deep Agents: LangChain's open-source agent harness

   According to the project README, Deep Agents is an open-source, opinionated agent harness for long-horizon, multi-step work. It includes sub-agents, filesystem access, context management, persistent memory, shell access, human-in-the-loop approval, skills and tools, while allowing components to be extended, overridden or replaced without forking. The project is model-agnostic and supports LLMs with tool calling, including frontier APIs, open-weight models and local models.

   GitHub Trending · Python · 🔥 7
2. 3d
   OpenWiki Evidence Graph Improves Agent Memory Consistency

   LangChain tested OpenWiki as a persistent memory layer for agents, finding in a vendor experiment that it was roughly five times more accurate than search alone on questions requiring information across multiple sources. Adding an evidence graph over wiki citations improved accuracy by a further 27%, doubled citation consistency and reduced cost per correct answer by 58%, though retrieval remained non-deterministic. For simple lookups, search without OpenWiki was equally accurate and faster, while reported answer costs excluded compilation and maintenance expenses.

   LangChain Blog · 🔥 0
3. 6d
   LangChain details five components of its company agent harness

   LangChain says internal agents for marketing, sales, and engineering converged on five shared primitives: working inside Slack, managing user or fixed identities, separating global and user memory, running code in sandboxes, and supporting scheduled or event-triggered runs. Its GTM Agent, OpenSWE, and Content Studio are built on Managed Deep Agents, which adds channels, cron jobs, sandboxes, memory, and identity infrastructure on top of the open-source Deep Agents harness. The architecture details are a first-party account, while the claimed impact on team workflows is not independently verified.

   LangChain Blog · 🔥 0
4. Oct 23
   LangChain Explores Context Engineering for Agent Retrieval

   LangChain’s blog examines three retrieval problems for agents with limited context windows: inefficient searching, excessive context returned at once, and irrelevant or stale information. It presents Deep Agents techniques including filesystem search-before-read tools, paginated file reads, offloading oversized results, delegating reading to subagents, and verifying grounding against retrieved sources.

   LangChain Blog · 🔥 0
5. Oct 21
   Agents CLI 1.8.0 adds first-party LangChain extension

   Google says Agents CLI 1.5.0 introduced an Extension System that lets developers add or override commands without forking the tool, while version 1.8.0 added a first-party LangChain extension. The developer blog demonstrates scaffolding, running, deploying, and evaluating a LangChain agent, plus a custom extension that wraps deployment with an SBOM scan; Google also claims the original ADK path can ship to Cloud Run with observability and evaluations wired up in about 30 minutes.

   Google Developers Blog · 🔥 0
6. Oct 14
   LangChain explains how Deep Agents use five techniques to manage context

   A LangChain blog post explains how its open-source agent harness Deep Agents addresses "context rot," where instructions, tool results and files accumulate during long tasks until they exceed the model's context window and degrade accuracy and reasoning. It describes five techniques: offloading large tool results to the filesystem, summarizing older messages as usage nears the limit, loading skills on demand, isolating noisy work in subagents, and storing long-term memory as a plain file. LangChain says that memory lives in storage the user controls, so the agent can switch models or providers while keeping what it has learned, and that thresholds and middleware behavior are configurable.

   LangChain Blog · 🔥 0
7. Oct 13
   LangChain adds ScaleDown context middleware for long-running agents

   LangChain's blog introduced the langchain-scaledown integration and said ScaleDown's constrained decoding models are now a direct model provider in the LangSmith LLM Gateway. The package ships with three middlewares for summarizing old history, compressing large tool results and extracting structured data, covering ScaleDown's four single-task models: Summarize, Compress, Extract and Classify. LangChain says each model accepts up to 1M tokens of input and that some handle images and scanned PDFs; ScaleDown's own benchmarks claim large speed and cost advantages over frontier models on narrow tasks, without independent verification in the post.

   LangChain Blog · 🔥 0

Experience and discussion from the community

Share my Deep Agents experienceAsk about Deep Agents