LangChain Expands Polly AI Assistant Across LangSmith Platform

Iris Coleman Mar 18, 2026 16:17

LangChain's Polly AI debugging assistant now available platform-wide in LangSmith, adding persistent memory and automated actions for AI agent developers.

LangChain Expands Polly AI Assistant Across LangSmith Platform

LangChain has rolled out its Polly AI assistant across the entire LangSmith platform, moving the debugging tool from limited placement to universal availability. The expansion comes just one day after the company announced a security patch for a critical vulnerability and days after unveiling an enterprise partnership with NVIDIA.

What Polly Actually Does

For developers building AI agents, debugging is a nightmare of complexity. Traces can run hundreds of steps deep, prompts stretch thousands of lines, and failures hide in the middle of sprawling execution chains. Polly is designed to parse all of that automatically.

The assistant can now read a 300-step trace, identify where things broke, and explain what happened. Previously confined to trace pages, thread views, and the playground, Polly now appears in the bottom-right corner of every LangSmith page—tracing projects, experiments, datasets, annotation queues, evaluators, you name it.

Three Changes That Matter

First, Polly maintains conversation context as developers navigate between pages. Start debugging a trace, jump to experiments for comparison, and the assistant remembers what you were working on. That sounds minor until you've lost your mental thread switching between views for the twentieth time.

Second, Polly can now take direct action. It'll update prompts, create datasets from failing runs, filter project views, write evaluator code, and compare experiments. It's shifted from answering questions to actually doing things.

Third, the assistant handles thread analysis for multi-turn conversations between users and agents. Instead of reading through every message manually, developers can ask whether users seemed frustrated, what issues they experienced, or whether their problems got solved.

Timing and Context

The release follows a busy week for LangChain. On March 17, the company launched LangSmith Sandboxes for secure AI agent code execution. The same day, security researchers disclosed vulnerabilities in LangSmith alongside Amazon Bedrock and SGLang that could enable data exfiltration. LangChain had previously patched a critical vulnerability (CVE-2026-25750) in December 2025 that exposed users to potential token theft.

On March 16, LangChain announced an enterprise agentic AI platform built with NVIDIA, signaling a push into larger enterprise deployments.

Getting Started

Existing LangSmith users can access Polly immediately via Cmd+I on Mac or Ctrl+I on Windows and Linux. The feature requires adding an API key for a model provider as a workspace secret—a two-minute setup according to LangChain's documentation.

LangSmith's pricing starts free at 5,000 traces monthly, with paid plans beginning at $39 per seat. The platform reached general availability with paid tiers in July 2024 after launching in closed beta in August 2023.

For teams already deep in agent development, the expanded Polly represents a meaningful workflow improvement. Whether it justifies switching platforms for those using alternatives remains the harder question.

Image source: Shutterstock