Simple, usage-based pricing. Models at cost.

One reasoning turn by one agent — typically one LLM call plus any tool invocations it triggers. A typical workflow runs 5–30 steps depending on complexity. Steps are metered consistently across all tiers.

Paid plans include a monthly step allotment, then bill any additional steps automatically at your tier's overage rate ($0.20/step on Pro, $0.15 on Business), shown live in your dashboard. The free Developer tier is a hard cap instead: runs pause when you hit the limit, so you never get a surprise bill.

Exactly what the provider charges, and nothing more. Every step records the provider's reported input, output and cached tokens (and, for web-search models, the per-request search fee), priced at that provider's published list rate. That amount is shown per step, per run and per model in your dashboard, and invoiced to paid plans at cost each period — 0% markup. On the free Developer plan, model usage is covered by your included steps.

The provider's list price, passed through. Training is billed once when a job finishes — the trained-token count the provider reports times its training rate (e.g. $5 per 1M tokens for GPT-4.1 mini, $25 for GPT-4.1, $2 for Mistral Small) — and the fine-tuned model is then billed per call at its own inference rate, shown next to every job. There is no platform fee for training; it's included in Pro and Business. Mistral additionally charges $2/month per stored model directly.

Yes — that's the point. Set a default model on the workflow and override it per agent: a Perplexity Sonar step to research the web with citations, a cheap fast model to classify, a frontier model to reason, another to review. Tool calling, approvals and checkpoints behave identically across Anthropic, OpenAI, Google, xAI, DeepSeek, Mistral and custom OpenAI-compatible endpoints.

You don't have to. LoopLlama's pooled provider access is included on every plan and billed at the provider's list price. Enterprise customers who prefer to pay providers directly can bring their own keys or run a self-hosted deployment with their own provider configuration.

Workflow inputs, outputs, and traces are stored encrypted at rest, scoped to your account, and never used to train models. Connector credentials are encrypted with a separate key. You can delete workflows and their run history at any time.

On the Enterprise plan we can run a single-tenant deployment in a dedicated VPC with the same API and SDK. Contact sales to scope it.

Yes. Upgrade, downgrade, or cancel from the dashboard at any time through the Stripe billing portal. Plans are billed monthly.