The Whole Platform Is Now Open Source: LangWatch May 2026 Update

May was a landmark month for us. The headline is simple: the entire LangWatch platform is now open source. On top of that, we brought voice testing straight into Scenario, graduated the rebuilt Traces experience to general availability, and opened up an all new AI Governance beta. Here is everything that shipped.

The Entire Platform Is Now Open Source Under Apache 2.0

This is the big one. Every core piece of LangWatch is now open source under the Apache 2.0 license: tracing, evaluations, LLM-as-judge, prompt management, datasets, experiments, the AI gateway, and Scenario. Nothing behind a wall, nothing held back.

You can stand up the complete stack with a single command:

npx @langwatch/server

That brings everything up on localhost:5560. From there you can trace your agents, catch odd behavior before it reaches users, keep costs in check through the gateway, and run simulations, all locally, all yours.

To be clear on what stays commercial: LangWatch Cloud and our enterprise capabilities remain paid. That covers SSO and SCIM, audit logs, advanced governance, and data retention. The open source project gives you the full engine; the commercial side is about running it at scale inside a company.

Voice Testing Is Now Native in Scenario

Scenario can now test full voice-to-voice conversations, not just text. You get simulated callers, scripts, traces, playback, and judge based evaluations that run against real voice output rather than a transcript stand-in.

The playback piece matters more than it sounds. When a call feels off, you can listen back and actually hear why instead of guessing from logs. It works across the voice stack you are already using: OpenAI Realtime, ElevenLabs, Gemini Live, Pipecat, Twilio, ComposableVoice, and plain WebSocket. And because it runs headlessly, you can wire voice tests into CI and catch regressions on every change.

The New Traces UI Is Now GA

The rebuilt Traces experience is out of beta and generally available. We rewrote it from the ground up for the messy reality of modern agents: multi-turn conversations, many tools, and multiple agents working together.

A few things we are proud of:

• Density modes. Save one lens and view it two ways. Engineers get a tight, data dense layout; PMs and domain experts get a more readable version of the same saved view.
• Seed live data before any code. New teammates can pull real data into the browser with a personal access token, so they can explore before writing a single line of SDK code.
• Ask AI, built in. When you hit an error or a confusing trace, Ask AI helps you dig in and understand what happened.

AI Governance Is Now Open for Beta

We opened the beta for AI Governance, which has two front doors: a personal home for developers and a governance home for admins.

Getting started is quick. langwatch login signs you in through your organization's SSO. From there, the commands you already run keep working exactly as before, whether that is langwatch claude, langwatch codex, langwatch cursor, or langwatch gemini. The only difference is that every request now flows through the gateway and gets attributed to a cost.

For developers, a personal /me portal shows the month's spend against your budget, your request counts, and a breakdown by model. Every span lands in your own Trace Explorer, so your usage is fully visible to you.

For admins, there is a wider view: organization wide spend, top spenders, and anomaly detection that flags spend spikes, geographic mismatches, and off hours activity. You also get ingestion source health and a complete audit log.

Every event is mapped to OCSF and can be replayed into Splunk, Datadog, Sentinel, or Elastic, so it slots into your existing security tooling. Governance reaches beyond LangWatch itself too, covering external platforms like Workato, Microsoft Copilot Studio, and the OpenAI and Anthropic compliance APIs. And every governance capability ships with a REST API, a CLI, and an MCP server, so you can automate it however you like.

Join Us Live

Two things coming up in June:

• On June 11, 2026 we are running a webinar on testing voice agents the same way you test chat agents.
• On June 17, 2026 we are hosting an in person event in Amsterdam during AMS Tech Week, together with our customer Altura, on LLM evals and agent simulations.

You can try all of this today. Run npx @langwatch/server to spin up the open source stack locally, head to app.langwatch.ai for the cloud, or read the docs at langwatch.ai/docs.

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