LangWatch Agent Observability and Evals

LangWatch B.V. builds an open-source agent observability and evaluation platform, and its site footer says it is built in Amsterdam. It addresses the difficulty of testing and governing LLM agents, from custom agents to coding assistants. Its blog reports a EUR 1 million funding round in February 2025 and a fully open-source release of the platform in May 2026.

The platform combines Simulate (multi-turn text and voice user simulations with a JudgeAgent), Evaluate (LLM-as-a-judge and online production evals), Observe (OpenTelemetry-native tracing with token and cost tracking) and an AI Gateway with per-team budgets. Langy, an automated AI engineer, turns test failures into pull requests. Named customers include Backbase, PagBank, Visma and Deloitte.

• Simulate: multi-turn text and voice user simulations scored by a JudgeAgent.
• Evaluate: LLM-as-a-judge, notebook and UI evals, and online production evals.
• Observe: OpenTelemetry-native tracing with token and cost tracking and topic clustering.
• Langy: an AI engineer that turns goals into test plans and failing tests into pull requests.
• AI Gateway: one endpoint for model calls with per-person, team and customer budgets.
• Prompt management, voice AI testing and Claude Code usage tracking.

A Vietnamese bank can simulate customer conversations against a virtual assistant before launch. A contact-centre operator can test voice agents in CI. An engineering team can track Claude Code token use across developers.

The platform is open source under Apache 2, so teams can self-host with Docker, Kubernetes or in their own VPC. Cloud regions cover the EU, US, UK and APAC. ISO 27001 certification and GDPR compliance with EU data residency are listed. Budgets per person, team or customer help control token spend.

Teams are shifting from manual prompt checks to simulated testing and continuous evaluation of agents, and are adding governance for coding assistants. Open-source, self-hostable tools with APAC regions fit Vietnamese firms that want control over data.

Legal entity LangWatch B.V., built in Amsterdam per the site footer. EUR 1 million funding round in February 2025; platform fully open sourced in May 2026; v3.0 in April 2026. ISO 27001 certified, monitored with Vanta.