The original title is "Harbor x LangChain: A Unified Stack for Evaluating Agents"

Short summary

Harbor is an open-source evaluation framework that runs AI agents in isolated, reproducible sandboxes—solving the limitations of traditional output-based evals for long-running, stateful agents. This LangChain video integrates Harbor with Deep Agents and LangSmith, walking through agent building, eval dataset structuring, and result tracking. Real environments with parallel execution enable deterministic, scalable testing.

• •Harbor runs agents in isolated sandboxes for reproducible, deterministic evaluation
• •Integrates with LangChain Deep Agents and LangSmith for full workflow observability
• •Replaces output-string evals with real environment testing—critical for complex, stateful agents

Generated with AI, which can make mistakes.