LangWatch

LangWatch is an open-source platform for engineering AI agents toward production. It is best suited to teams that need simulation, evaluation, and production tracing across one workflow. Its deployment flexibility is a strength; small free-plan quotas make it a limited fit for sustained team use.

Overview

LangWatch brings agent simulations, evaluations, and observability together, covering work before launch and on production traffic. Simulations exercise multi-turn conversations, tool use, recovery, and business outcomes; evaluations can run on datasets, production traffic, or in CI/CD as merge gates. Production traces report agent steps, cost, tokens, duration, errors, evaluations, and annotations, making the platform relevant to teams that need both behavior checks and operational visibility.

The product is open source under Apache 2.0 and can run as managed cloud, self-hosted, or hybrid. Self-hosting uses Docker Compose and the customer's ClickHouse, so it offers deployment control but also assumes infrastructure ownership. LangWatch B.V. is based in Amsterdam.

Key features

• Agent simulations: Test conversations, tool use, recovery, and business outcomes before launch. This is useful when an agent's behavior depends on several turns or tool calls, rather than a single response.
• Evaluations: Run offline dataset checks, continuous production evaluations, or CI/CD gates. Developers, product owners, QA, and domain experts can use these workflows at different stages of delivery.
• Tracing and cost visibility: Agent traces expose steps, token and cost usage, duration, errors, evaluations, and annotations. This combines debugging signals with spend tracking, though the free plan limits event volume and data access.
• Integrations and voice testing: Python and TypeScript SDKs, OpenTelemetry, REST ingestion, model providers, and frameworks such as LangGraph, LangChain, CrewAI, and DSPy provide several connection paths. Voice testing supports OpenAI Realtime, Gemini, ElevenLabs, Google Live API, Pipecat, WebSocket bots, and Twilio.
• Security and privacy: The trust center lists AES-256 encryption at rest, TLS 1.2+ in transit, AWS KMS, RBAC, MFA, and SSO options. It also describes automatic PII detection and removal, GDPR right-to-be-forgotten, and enterprise DPAs.

Pricing

LangWatch is freemium. The Developer plan is free, billed free forever. It includes 50k events/month, 14-day data access, 2 users, 3 scenarios, 3 simulations, 3 custom evals, and community support through GitHub and Discord. It is a sensible starting point for a small team exploring workflows, but the event, access, and evaluation caps constrain ongoing production use.

Growth costs €29 a month, billed per core-seat / month. It includes 200k events, with additional events at €5 / 100k, 30-day retention, extendable at €3 / GB, unlimited lite-users, simulations, evals, and prompts, plus private Slack / Teams support. This removes the free tier's simulation and evaluation ceilings for teams that need regular iteration; event volume and longer retention can add cost.

Enterprise has custom pricing. It offers hybrid, self-hosted, or on-prem deployment; custom retention and SSO/RBAC; audit logs and SLAs; ISO 27001 reports, InfoSec and legal review; and custom terms and DPA. It is aimed at organizations with deployment, governance, or contractual requirements beyond Growth.

Platforms

LangWatch supports API, Linux, self-hosted, and web. Managed cloud, self-hosted, and hybrid deployment give teams a choice between managed operation and infrastructure control, with self-hosting requiring Docker Compose and ClickHouse.

Who it's for

LangWatch fits AI teams shipping agents that need to test multi-turn behavior before release and monitor performance after launch. Its evaluation workflows also suit developers, product owners, QA, and domain experts. Teams with strict deployment or governance needs may benefit from Enterprise; individuals or small teams with modest evaluation needs can begin on Developer.

Pros and cons

• Pro: Simulations, offline and live evaluations, and CI/CD gates cover several stages of agent development.
• Pro: Cost, token, duration, error, and step traces make agent behavior and operational signals visible together.
• Pro: Managed, self-hosted, and hybrid deployment accommodate different infrastructure preferences.
• Con: Developer's 50k monthly events, 14-day access, and three-item simulation and evaluation limits can quickly constrain a growing workflow.
• Con: Growth adds event and retention charges beyond its included amounts, so heavier use can cost more than the core-seat price.
• Con: Self-hosting depends on Docker Compose and customer-managed ClickHouse rather than being a turnkey deployment.

Alternatives

For broader comparisons, browse AI LLM Observability Tools, Model Monitoring Software, LLM Observability Tools, AI LLM Evaluation Tools, AI Agent Evaluation Tools, and AI Agent Observability Tools.

• OpenLIT is a free alternative with unlimited self-hosted usage, users, projects, and environments on its OSS plan; choose it if those uncapped self-hosted allowances matter more than LangWatch's stated agent-simulation and voice-testing workflows.
• W&B Weave offers a free plan with 1 GB/mo ingestion, 5 GB/mo storage, and evaluations, tracing, and scorers; it suits readers who want those explicit free-tier allowances.
• Opik offers a free cloud plan and an open-source core users can run locally; consider it when running the core observability and evaluation set locally is the priority.
• Agenta offers a free Hobby plan with 2 team members, 5,000 agent runs/month, 20 evaluations/month, and 1-week trace retention; it is an option for a small team whose needs fit those caps.
• Arize AX has a free SaaS plan with 25k trace spans/month, 1 GB ingestion/month, 15-day retention, and unlimited users and evaluations; choose it if those free-tier limits suit the workload.
• Arize Phoenix is a free, open-source, self-hosted alternative.
• Braintrust has a free Starter plan with 1 GB processed data, 10,000 scores, 14-day retention, unlimited users, projects, and datasets; it may suit teams prioritizing those allowances.
• Confident AI offers a free plan with 2 user seats, 1 project, 5 test runs per week, and 1 GB-month of trace spans; it is an option for readers whose testing needs fit that scale.

Verdict

Choose LangWatch if your AI team needs agent-specific simulation, evaluation, and production tracing, with a choice of managed or controlled deployment. Its strongest case is the joined-up workflow from pre-launch tests to production signals. Look elsewhere if Developer's quotas are too tight and Growth's per-seat, event, and retention costs or self-hosting requirements do not fit your team.