正在加载视频...
视频加载失败
Today, we’re shipping new ways to observe, analyze, and debug agents with LangSmith: • Polly: an AI assistant for AI engineering that helps you understand traces, threads, and improve prompts • LangSmith Fetch: a CLI for pulling trace & thread data straight into your terminal or coding agent Agents... show more
11 条评论
this looks amazing! wonder what these "reading file {file_name}" logs are, as seen both in threads and in prompts are these part of a kind of sandbox file dir the agent has access to? also the threads/trajectories seem to be saved and read as md files, really curious how this works under the hood! cc @hwchase17
Real Q: Is fetch output optimized for agent context windows (like Cursor) or is it just raw JSON ? Piping full traces usually = instant token overflow Curious if the CLI formats it to actually fit the token budget
Cool! I've doing this kind of stuff externally with the Langsmith MCP, this is far better! Thanks team!
I see so many questions in Slack that leaning on Polly would solve. So glad we built this!
Looks like a solid step forward for debugging in complex agent scenarios. Curious about how Polly integrates with existing workflows-any info on ease of adoption?
Sweet, now I can build a triage agent to watch over the agent that's looking over my agent observability platform
Why no MCP tool for langsmith fetch? So any AI code agent can directly use it.
This is great
, This hits a real pain point for anyone building serious agents. Once agents start running longer and branching into tools, memory, and retries, debugging turns into archaeology. Traces everywhere, failures that don’t reproduce, prompts that “worked yesterday.” Polly feels like a missing layer for understanding what’s actually happening, not just guessing. LangSmith Fetch is just as important. Pulling trace data straight into the terminal and your coding flow is how this should work. Debugging agents inside a dashboard alone never scaled. Devs live in terminals and editors, not tabs. The shift is from “prompting models” to engineering systems. And systems demand observability, not vibes. Tools like this are what take agents from cool experiments to software you can trust in production. If you’re reading this, you probably care about AI that’s actually useful to you. Check out @techificial. We share practical tools, real case studies, and resources you can apply immediately. Follow us and start with the pinned post to see what we’re building.
Exciting updates! Full visibility into agent behavior is essential as autonomous agents get more complex. Tools like Polly and LangSmith Fetch are exactly what teams need to debug and reason over agent workflows effectively.
this is awesome, can’t wait to use it love the name
