langchain==1.3.15 リリースlangchain==1.3.15

LangChain has released version 1.3.15 of its core Python library, an incremental update to one of the most widely adopted frameworks for building applications and autonomous agents on top of large language models. While the release is a routine maintenance step rather than a major feature drop, it introduces a new configuration surface for agent tracing that is likely to interest teams running LangChain-based agents in production.

The most notable change is the exposure of trace_policy on AgentMiddleware. In LangChain's agent architecture, middleware provides a structured way to intercept and augment the steps an agent takes, for example wrapping model calls, tool invocations, and state transitions with custom logic. By surfacing trace_policy as a configurable property on AgentMiddleware, the release appears to give developers more direct control over how those steps are traced and recorded, rather than relying solely on defaults. This kind of control is often used to manage trace verbosity, scope what is captured, or align an agent's behavior with an organization's observability and data-handling requirements.

Tracing has become a central concern as agentic systems grow more complex. A single agent run can involve multiple model calls, tool executions, and branching decisions, which makes visibility into each step important for debugging, evaluation, and cost tracking. LangChain typically pairs with LangSmith, the company's platform for tracing, evaluating, and monitoring LLM applications, and with LangGraph, its library for building stateful, graph-based agent workflows. A more explicit trace_policy hook fits into that broader push toward giving developers finer-grained instrumentation without forcing them to rewrite their agent logic.

The release also includes housekeeping changes. One entry fixes type errors in the test suite, part of ongoing efforts to keep the codebase type-safe and to reduce friction for contributors who rely on static analysis tools. These changes do not alter runtime behavior but help maintain code quality as the library evolves.

On the dependency side, the release bumps h2 from 4.3.0 to 4.4.1 in the langchain_v1 library path. h2 is a pure-Python implementation of the HTTP/2 protocol used lower in the networking stack, and dependency bumps of this kind are commonly driven by upstream bug fixes or security advisories. Keeping transitive dependencies current is a routine but important part of maintaining a package that sits at the base of many production deployments. The changelog lists addit