Agent Atlas / Agent frameworks

langchain-ai/langchain

A broad agent application framework whose middleware, tool, model, retry, interrupt, and persistence seams expose how application code shapes an agent loop.

Agent framework Scope profile Reported + sourced Reviewed July 21, 2026

Intended operator

Developers building agent applications and workflows.

Primary surfaces

Python and JavaScript libraries / agent and tool APIs / middleware

Tracked repository

langchain-ai/langchain

Teaching mechanism

Interrupts must survive retry policy

Control-flow interrupts are separated from ordinary failures so a retry layer does not consume the operator's stop or handoff signal.

Evidence limitA framework-level distinction still depends on each application wiring the control path correctly.

Inspect the source

System map status

Reporting is still in progress.

This scope profile names the project's place in the field and one useful mechanism. It does not claim that every runtime boundary has been inspected.

Learn Through This Project

7 lessons

Chapter 1 Before the First Model Request Chapter 3 Tool Calls: Contract, Authorization, Execution Chapter 4 Loops, Workflows, and Recovery Chapter 6 Memory Is a Typed Storage Contract Chapter 7 The Approval Trap Chapter 9 Trust Must Make Its Case Chapter 10 The Future Has Three Branches

Compare Its Boundaries

6 labs

Comparison labTool Contracts Comparison labGoverned Memory Comparison labApproval Boundaries Comparison labSessions And Recovery Comparison labEvidence And Evaluation Comparison labMaturity Signals

Latest Reporting

8 articles

API Discovery: One Reasoning-Effort Dial Across Provider Integrations

July 21, 2026 — LangChain's new reasoning_effort interface lets a builder express an effort choice once across supported chat integrations—provided the selected model profile says that choice exists. langchain-ai/langchain

LangChain Stops Retries From Eating Interrupts

July 9, 2026 — LangChain's retry middleware repair is a people/context story because the project's public human-in-the-loop work makes one small bug fix read as a guardrail for users trying to stop an agent. langchain-ai/langchain

Model Catalogs Get Plain-English Receipts

June 24, 2026 — LangChain and Pi are making model/provider changes easier to inspect before they become hidden assumptions inside agent runtimes. langchain-ai/langchain, earendil-works/pi

Agent Tools Are Moving Safety Into Boundaries

June 16, 2026 — Gemini CLI, LangChain, and Codex landed fresh patches that put agent safety in filesystem checks, type gates, pending-tool waits, and permission-path parsing. google-gemini/gemini-cli, langchain-ai/langchain, openai/codex

Agent Tooling Gets Runtime Guardrails

June 11, 2026 — Fresh OpenCode, LangChain, Gemini CLI, Codex, and OpenClaw commits show agent projects tightening the machinery around tools, sessions, streams, and approvals. anomalyco/opencode, langchain-ai/langchain, google-gemini/gemini-cli, openai/codex, openclaw/openclaw

Agent Failure States Are Becoming Instructions

June 3, 2026 — Fresh LangChain, Crush, Gemini CLI, and OpenClaw commits show agent tools turning denial, fallback, routing, and policy drift into explicit next steps. openai/codex, google-gemini/gemini-cli, openclaw/openclaw, charmbracelet/crush, langchain-ai/langchain

Agent Reliability Is Moving Into the Test Rig

May 25, 2026 — The newest source-readable trail points away from model mystique and toward the workflows that test, trace, serialize, and maintain agent behavior. google-gemini/gemini-cli, langchain-ai/langchain

Agent Runtimes Are Learning Where to Say No

May 24, 2026 — After shipping more visible control planes, agent projects are now tightening the places where tools, plugins, credentials, chat context, and test runs are allowed to flow. openai/codex, langchain-ai/langchain, NousResearch/hermes-agent

Primary Sources

LangChain repository

Primary framework source.

Open Questions

Which runtime guarantees belong to the framework, and which remain application responsibilities?

Evidence state describes what this Atlas profile can support today. It is not a product score and it is not evidence that uninspected capabilities are absent.