Pricing

Golem Cloud pricing

Run your agents and tools on infrastructure we manage. Start with the free Developer Preview.

Planned
Rates to come

paid service is not available yet

Planning a production deployment? Talk with us about your workload and requirements.

  • Planned pricing for compute, memory, and filesystem usage
  • Launch date and commercial terms to be announced
  • Discuss capacity, deployment, and support needs

A dollar estimate needs the rates and terms for paid Cloud.

Discuss your workload
The planned pricing model

Four resources. Measured separately.

Golem 1.6 measures execution work, active memory, and active filesystem space. These are the resource dimensions planned for paid Cloud; rates are not yet published.

Compute GCU
WebAssembly execution work. One Golem Compute Unit is one million fuel units—not a request count or a model token.
Active memory GiB-seconds
Allocated agent memory multiplied by the time the agent is active.
Active durable filesystem GiB-months
Allocated filesystem space for durable agents, accumulated while they are active.
Active ephemeral filesystem GiB-months
Allocated scratch space for ephemeral agents, accumulated while they are active.

While working or waiting actively

Execution consumes compute. Allocated memory and filesystem space continue to accrue usage while an agent remains active, including during an external wait.

While idle or suspended

These four resource meters stop accruing. An agent’s age is not its active time. This does not promise unlimited data retention.

Plan your workload

Start with usage, then work out cost.

An agent run has no fixed resource cost. Measure a representative workload before extrapolating to a month of traffic.

  1. Try a representative workload

    Exercise the code paths and payload sizes you expect to run. Separate active work from time spent suspended.

  2. Gather the quantities

    Record GCU, active memory, and durable and ephemeral filesystem usage. Include expected monthly volume and peak capacity.

  3. Discuss the production requirements

    Share your measurements, launch timing, retention needs, and support or private-deployment requirements. Rates and paid availability remain unannounced.

Golem Cloud runs your agent code and tools. Budget separately for model APIs and other services billed through your provider accounts.

Need to run Golem yourself?

Self-host the runtime

Operate Golem on your own infrastructure. You supply the compute, storage, and operations. The repository contains the software license and setup instructions.

View on GitHub

Plan a private deployment

The packaged Cloud On-Prem offer is planned. Discuss your cloud, Kubernetes, support, and security requirements with us.

Discuss private deployment

Pricing questions

Is the Preview a permanent free tier?

No. The managed service is free during Developer Preview for evaluation and prototyping. A permanent free production allowance has not been announced. Paid availability, rates, and included quantities will be published when defined.

What happens when I reach a resource limit?

The consequence depends on the limit. In Golem 1.6, configured monthly compute or call limits can suspend durable agents and fail ephemeral execution. Per-agent memory or managed-filesystem limits can reject allocations or writes. Account limits can reject creation of additional resources. Contact us to confirm the limits that apply to your Cloud workload.

Can resource limits set a maximum monthly bill?

No. Resource limits and application quotas are not a whole-bill spending cap. In the 1.6 metering model, monthly memory and filesystem usage are measured without a corresponding monthly runtime cap. Application quotas govern operations your code explicitly reserves; they do not automatically cap model-provider invoices.

How do the memory and filesystem units work?

A GiB is 1,073,741,824 bytes. Memory is allocated GiB multiplied by active seconds. Filesystem usage is allocated space multiplied by active time, normalized to a 730-hour month. The 1.6 usage API labels these binary-based quantities GB-seconds and GB-month. Usage is reported by UTC calendar month.

What do I need before planning a production launch?

Paid Golem Cloud is not available yet. Discuss your expected volume, measured resource usage, peak capacity, retention, and support requirements with us. A production budget will need published rates and commercial terms. Preview has no SLA or retention guarantee and must not be used for production workloads.

Try your agents on Golem Cloud.

Deploy a prototype, exercise its real workload, and see how it behaves.