The LangWatch Monthly Drop: January 2026

We kicked off 2026 with one of our busiest release cycles yet. This month is less about small refinements and more about reshaping how teams build, test, and ship reliable agents inside LangWatch. Here is everything that landed in January.

A fresh look for LangWatch in 2026

We started the year with a ground up makeover of the LangWatch experience. The biggest win is how much easier it now is to build Evals directly in the UI, without wrestling with the interface to get there.

We designed the new experience so that everyone who touches agent quality feels at home in it. Whether you are an engineer wiring up pipelines, a PM shaping what good output looks like, or a domain expert who knows the edge cases better than anyone, the redesigned product is meant to be a pleasure to work in. Log in to app.langwatch.ai and take it for a spin.

Scenario grows from code into the platform

Agent simulations have always been one of the most powerful things you can do with LangWatch, but until now they mostly lived in code. That has changed. Scenarios are now a first class part of the platform.

You can build lifelike user situations, spell out exactly what counts as a pass or a fail, and run complete end to end agent tests against either a prompt or a live HTTP agent. No bespoke test harness required.

The point is that the whole team can pull in the same direction:

• PMs describe what they expect the agent to do
• Domain experts contribute the tricky edge cases they know from the field
• Engineers get full traces so they can debug every single step

AI assisted generation makes spinning up new scenarios quick, and with saved history plus one click re runs, agent testing becomes a continuous quality loop rather than a one off. You can fire off thousands of simulations both before and after a release to see exactly what moved. To try it out, go to Scenarios in the platform.

Evaluations v3

With v3 we focused on pulling evaluations tighter into the everyday workflows of developers and PMs.

Two changes stand out. First, prompt management got a redesign so that prompts are now first class, fully versioned, and wired directly to your evals. Second, evaluations are now replicable: you can rerun the same logic across different models, prompts, or agent versions, either from the platform or straight from your CI/CD.

The bigger goal is to fold evals into the product development lifecycle itself, so PMs and domain experts can meaningfully contribute without stripping any control away from the developers who own the pipeline.

Alerts on your dashboards

A dashboard that only tells you something is wrong when you happen to be looking at it is only doing half the job. So we added alerting to custom dashboards.

Now your team can be notified the moment a regression appears, quality dips, or the agent starts behaving in ways you did not expect. Instead of passively watching a chart, you get pinged when performance drops so you can react right away.

Now on the Azure and Google Cloud Marketplaces

LangWatch is now listed on both the Azure Marketplace and the Google Cloud Marketplace. That makes procurement, security reviews, and deployment inside your own cloud environment far simpler to navigate. Private offers are available too, so reach out to us if that fits your setup.

Looking ahead

Next up, we are going deeper on agent simulations, and we are opening a first beta to simulate voice agents. The last few spots are filling up, so contact us soon if you want in.

Also shipped

We also rolled out LangWatch v2.0, which packs in a stack of improvements:

• Custom dashboards with drag and drop plus resizable graphs
• Database backed management for agents and evaluators
• A keyword subfield on metadata.name for sorting
• Replicable evaluations
• An upgraded model registry with dynamic parameters and OpenRouter sync

You can dig into the details anytime in the docs.

LangWatch is the open source LLM evaluation and agent testing platform.