# LangSmith Engine

## Your proactive agent engineer

LangSmith Engine analyzes production traces, groups related failures, and recommends fixes so your team can improve agent quality faster.

## Helping top teams ship great agents

### Three ways to build, from full harness to full control

Each package solves a different problem. Choose where you want to start, then compose the rest of the stack around it.

### Detect

#### Automatically detect issues in production traces

Engine analyzes LangSmith traces to look for patterns in unmet user expectations or agent failures. Based on your preferences, Engine will triage high severity issues, so you can understand where there’s opportunity for improvement or poor performance that needs to be addressed.

Engine can:

- **Run** **_automatically in the background_**
- **Cluster** _related traces into issues_
- **Prioritize** _issues based on severity and your preferences_

[Set up Engine](https://docs.langchain.com/langsmith/engine#set-up-engine-for-a-tracing-project)

### Fix

#### Engine writes the fix

For each issue, Engine summarizes the failure mode, identifies what needs to change, and writes the prompt or code fix. If you connect your codebase, Engine can open a GitHub PR with the proposed change, ready for your team to review and merge.

Engine can:

- **Write** _prompt and code changes based on production failures_
- **Show** _diffs with explanations_
- **Open** _GitHub PRs for review_

[Configure Engine to write PRs](https://docs.langchain.com/langsmith/engine#open-a-pull-request)

### Prevent

#### Prevent issues from coming back

Like a good engineer, Engine writes tests so mistakes don’t reappear. It suggests online evaluators to track whether the issue stays resolved, and recommends examples to add to your offline eval datasets so you can catch regressions as you evolve your agent.

Engine can:

- **Suggest** _online evals for production monitoring_
- **Recommend** _examples for offline experiments_
- **Build** _a feedback loop from traces to fixes to evals_

[Create evals automatically](https://docs.langchain.com/langsmith/engine#take-action-on-an-issue)

### Resources for LangSmith Engine

[Get started with LangSmith Engine](https://docs.langchain.com/langsmith/engine)  
[Introducing LangSmith Engine](/content/blog/introducing-langsmith-engine/index.html)  
[How We Built LangSmith Engine | Interrupt 26](https://www.youtube.com/watch?v=MVvFDfxaeWg)

### FAQs for LangSmith Engine

**What model do you use for LangSmith Engine?**

LangSmith Engine uses LLMs provided through LangChain. We do not support bringing your own model for Engine so that we can control the full Engine experience and deliver the best results for customers. Please read more about our subprocessors [here](https://trust.langchain.com/subprocessors) and our notice to customers [here](https://trust.langchain.com/updates).

**What can I share with our security team on using Engine?**

All model providers operate under zero data retention and are contractually prohibited from training or fine-tuning on your data.

**How is Engine priced?**

LangSmith Engine is a standalone agent that consumes LangChain Compute Units while it works. LCUs are a normalized unit of work that account for compute, storage, memory, and LLM usage.

Usage depends on the number of traces Engine analyzes, the depth of analysis required, and the amount of work it performs. To learn more about pricing, visit our [pricing page.](/content/pricing/index.html)

### Ready to add an active agent engineer to your team?

[Start building](https://smith.langchain.com/) [Get a demo](/content/contact-sales/index.html)