Ai

LangChain Perplexity Integration Adds Responses API Flag

LangChain Perplexity Integration Adds Responses API Flag

LangChain dropped version 1.3.0 of its Perplexity integration, and it's not just a routine maintenance release. The headline feature is a new use_responses_api flag on ChatPerplexity. This lets developers toggle between the chat completions endpoint and Perplexity's newer responses API—a shift that gives more direct control over the model's output behavior.

What Changed

The core addition is use_responses_api, a boolean parameter on the chat model class. When set to True, the client routes requests through the responses endpoint rather than the default chat completions. That sounds simple, but it unlocks capabilities like streaming raw response objects and accessing metadata not available in the older endpoint.

Beyond that, the release bumps several dependencies. LangSmith updated from 0.8.0 to 0.8.5, and the idna library jumped from 3.10 to 3.15—a security patch for internationalized domain names. The team also tightened the langchain-tests floor to version 1.1.9 and hardened Dependabot's version-bound preservation in CI. These aren't flashy, but they reduce the likelihood of breaking changes sneaking in.

Why It Matters

The use_responses_api flag is the real story here. Perplexity's responses API returns structured data including citations and confidence scores, which can be critical for enterprise apps that need to audit AI outputs. For developers building search-summary tools or fact-checking workflows, this is huge.

But there's a catch: the flag is opt-in. That means existing code won't break, but teams that want the new capabilities have to explicitly enable it. It's a conservative rollout, which I think is smart—let early adopters test the waters before making it default.

One minor gripe: the changelog doesn't specify whether the responses API supports all models or just certain ones. Developers will need to check Perplexity's docs, but that's typical for edge features like this. Overall, it's a well-timed update as more teams demand transparency from AI providers. If you're running LangChain with Perplexity, give it a spin—the flag is worth exploring.

Official Source: https://github.com/langchain-ai/langchain/releases/tag/langchain-perplexity%3D%3D1.3.0