Pricing

How LangWatch Cloud pricing works, with events, data retention, and answers to common questions.

LangWatch Cloud pricing is built around two ideas: seats scale with your team, and usage stays cheap. Building reliable AI systems is collaborative work: engineers iterate, product teams define success metrics, domain experts review outputs, and stakeholders track performance. We price around that collaboration and keep usage costs low, so more tracking, evaluations, and production traffic adds little to your bill.

Start for free, no credit card required

The free plan gives you immediate access to all core features, including evaluations, monitoring, and agent simulations.

​

Events

LangWatch Cloud pricing is based on the number of ingested events per billing period.

An event is a standard telemetry point, the building block of everything you observe in LangWatch. The most common one is a span: a single step in your system, such as an LLM call, a tool execution, a retrieval step, a span timer firing, or the point at which an error occurred.

A billable event is any of these:

• A span, log or metric your app sent us. The telemetry that makes up a request to your agent or LLM application, or to a coding assistant you track. Spans are the individual steps; logs and metrics are what those steps recorded along the way.
• An evaluation result. One score from an evaluator, whether it ran on live traffic or over a dataset.
• A step of an experiment run. A sweep counts its start, each dataset row it ran, and each score it produced, so one experiment is many events rather than one.
• A step of a scenario simulation. A run counts its start plus each turn of the conversation.

Spans dominate the count whenever your app sends traces, but testing produces billable events too: an experiment over 100 rows with 2 evaluators is roughly 300 events. There are no separate charges per feature on top of this, and everything you create (prompts, evaluators, datasets, dashboards) is free.

For the exact arithmetic per family, what is never billed, and how retries are deduplicated, see What counts as a billable event.

​

What’s included

Each plan includes a base volume of events stored for a default window. For example, the Growth plan includes 200,000 events stored for 49 days. If you need a longer retention window or a higher volume, you can extend retention with an Enterprise plan.

​

Instant Evals

Instant Evals are priced separately from events. A judged query or run is charged per input token sent to the judge, at $0.0546 per million input tokens. Output is free. Judging ten thousand conversations of average size costs about a quarter of a dollar.

Input tokens are the text the judge reads plus the questions asked about it. One request carries every question about one text, so a statement with three eval columns costs about the same as one with a single column.

POST /api/v1/instant-evals/estimate returns the price of a run before it starts, from the rows the statement matches and a sample of their size. See Running an Instant Eval.

Organizations without a paid plan have a $1 free budget for Instant Evals in total, across all their projects. Past it, a new run or a judged query is refused until the organization upgrades. Paid plans have no such limit and are billed monthly for what was judged.

Every judged query and run is one spend event on the billing events ledger, so a gateway budget set on a project or an organization caps Instant Evals the same way it caps gateway traffic.

​

Custom data retention

Retention controls how long your events are kept before they’re automatically removed. Every plan keeps events for 49 days by default. Paid plans change the window in Settings → Data Retention, per organization, team or project:

Plan	Default retention	Windows you can set
Free	49 days	None. Trace content older than 14 days is redacted until you upgrade
Accelerate (legacy)	49 days	35 or 63 days
Growth	49 days	35 or 63 days
Enterprise	49 days	35 or 63 days, or any whole number of weeks from 49 days up

Retention is set in whole weeks.

​

Extending retention

Windows longer than 63 days are available on Enterprise plans. Extended storage is priced from €3 per GB and configured as part of your agreement, so you only pay for the additional storage you actually use.

Self-hosted deployments keep the same 49-day default. An Enterprise license opens the same windows as the Enterprise plan, and a platform administrator can then keep data indefinitely. See Editions and licensing.

To extend retention beyond 63 days, reach out to your contact or [e-mail].

​

Frequently asked questions

How does pricing behave as I scale up?

There are no sudden pricing jumps tied to volume. LangWatch is designed so teams can track all agent activity, run continuous evaluations, and scale to production workloads without their bill spiking. Seats scale with team size, and event pricing stays low, so a growing workload adds little to your bill.

Is there a free plan?

Yes. You can start for free with LangWatch Cloud, no credit card required. The free plan gives you immediate access to all core features: evaluations, monitoring, and agent simulations. It’s ideal for developers who want to experiment or run small-scale testing.

What if I'm working with sensitive data from the start?

Run LangWatch locally on your own machine. All features are available from our open-source repository, so you can build with sensitive data without it leaving your environment.

What is the difference between a full member and a lite member?

Which seat someone occupies follows from what they can do. A full member can change things: build prompts, run experiments, edit datasets, configure the gateway. A lite member can only look at what the team produces. They open the projects they are invited to and read traces, analytics, evaluations, scenario runs, datasets, prompts and experiments, and they can leave annotations on what they read. They cannot see costs, and they cannot create, edit or delete anything else.

Those are permissions rather than screens, so the same limits apply wherever they reach the data, including the API and the MCP server. Read-only reviewers, stakeholders watching dashboards, and anyone who comments but does not build are lite members. See Seats for the full breakdown.

What counts as a billable event?

Billable events are the spans, logs and metrics your app or a tracked coding assistant sends, evaluation results, the steps of an experiment run, and the steps of a scenario simulation. Pricing is based on the total number of ingested events per billing period. The full list is here, including how many events a sweep or a simulation produces, what is never billed, and how retries are deduplicated.

Can I customize how long my data is retained?

On LangWatch Cloud every plan keeps events for 49 days by default. Paid plans set 35 or 63 days in Settings → Data Retention, per organization, team or project. Enterprise plans set any whole number of weeks from 49 days up, and a higher volume of stored events. Reach out to [e-mail] to set up an Enterprise window or extra volume.

Self-hosted, the data is in your own database and keeps the same 49-day default. A license adds per-scope retention policies, with a separate window for traces, scenarios and experiments.

Are webhooks and the billing events API included in my plan?

The webhook endpoints platform (signed event delivery to your systems) and the billing events surface (per-request spend events, reconciliation pulls, replay) are Enterprise capabilities, on Cloud and on self-hosted licenses alike. They exist for platforms that rebill their own customers and need billing-grade delivery guarantees. Spend records are retained for a fixed 13 months independent of your data-retention settings. Reach out to [e-mail] to enable them.

Can I self-host or run a hybrid setup?

Yes. LangWatch is available both self-hosted and hybrid. With the hybrid option you get the convenience of running on our Cloud while your data stays on your side. See the self-hosting docs or reach out for more information.

You can run the self-hosted version for free, with unlimited users, projects, and events on your own hardware. If you require enterprise features such as SSO, RBAC, SCIM, gateway webhooks, or a support SLA, reach out to us. See Editions and licensing for what a license adds.

Where is my data stored?

For LangWatch Cloud, data is stored securely in the EU. For the US, Pacific, and other regions, contact us at [e-mail] and we’ll meet your requirements. For Enterprise users, we offer on-premise options so your data stays entirely within your own infrastructure.

Do you support custom contracts and procurement reviews?

Yes. We know procurement and legal teams often have their own review processes. For Enterprise and regulated-industry plans, we support custom contracting and custom workflows.

Is LangWatch secure and compliant?

LangWatch is ISO 27001 and GDPR compliant. Enterprise customers can request our security documentation, pentest results, and vendor assessments. For more, see langwatch.ai/security.

Can my security team review your security documentation?

Yes. Your security team can access the full set of LangWatch security documentation, including:

• Pentest report, performed by a CREST-certified third-party security firm, covering LangWatch Cloud and hybrid setups.
• Data flow & architecture diagrams, including deployment models, encryption protocols (in-transit and at-rest), and data boundaries.
• Latest ISO 27001 reports, covering more than SOC 2.

Reach out to our team at [e-mail].