Free to self-host Cloud when you need more

One predict call. Each item in a batch predict counts as one decision. Rewards, reads and campaign management are free.

A decision that spends from a limited budget: a discount or free-shipping offer, a callback from a human agent, a debt settlement offer, a retention credit, or a paid LLM or API call. BanditDB paces spend across decisions so the campaign keeps learning without going over budget. Budgeted decisions cost $6 per 1K, regular decisions $0.30 per 1K.

$60 of free usage on Cloud while it is in beta. That is about 10K budgeted decisions, 200K regular decisions, or any mix of the two.

Yes. Each Cloud account runs up to 10 campaigns at 10 requests per second, with up to 64 context features per decision, and outcomes are recorded when they arrive within an hour of the decision. Large neural models do not fit the beta size. If you need more, talk to us. Self-hosted has none of these limits.

There are no caps on decisions, campaigns or models. Every bandit algorithm, the REST API, the Python and JS SDKs, MCP and Parquet export are included. Cloud and Enterprise add the budget dashboard and alerts, the built-in AI agent, causal impact reports and team roles, and we run the servers for you.

Yes, in both directions. It is the same engine and the same API, so only the endpoint in your code changes. Export to Parquet to take your data with you.

No. Self-host from the docs, or start on Cloud with $60 of beta credits. The MCP server lets your AI assistant create and inspect campaigns for you. On Cloud, the built-in AI agent sets up your first decision flow from your website.

When BanditDB has to run in your own VPC or on-premises, when data residency is written into a contract, when you need a security review and SSO, or when you want a Forward Deployed Engineer on call, or need more than the Cloud beta limits. Enterprise is invoiced annually.