Breyta.ai

Breyta's command-line interface is the main way you build. You install it, hand your coding agent the setup prompt, and it produces a structured flow definition you can inspect before anything runs. Every command returns stable JSON, so an agent can parse results and act on them without a human in the loop.

Breyta runs your logic as a versioned workflow with triggers, state, retries, approvals, and logs. The deterministic runtime lets it recover from a failed step and repair a run safely instead of starting over. You get run history and failure details, which matters when a customer is waiting on the output.

Flows aren't just LLM calls. Breyta supports HTTP, code and function steps, database, search, wait, notify, key-value, sleep, SSH, and child workflows. That range means you can connect to almost any system with an HTTP step and use code steps for tricky auth or data mapping.

You can package the same workflow as a manual app, an HTTP endpoint, an MCP tool, or a reusable workflow step. That covers HTTP endpoints for developers, MCP tools for AI clients, and manual apps for everyone else. Buyers then use it however fits their stack. Breyta handles the app page, required connections, example outputs, and the checkout flow, and creators keep 95% of marketplace sales.

Workflows can bind your own database, blob storage, key-value store, HTTP APIs, and model providers through managed connections instead of forcing data into Breyta's storage. Secrets sit in KMS-backed encryption and stay separate from the workflow definition, so credentials never get hardcoded into a flow.

Billing runs on step executions rather than seats or workflow counts. Every paid plan includes unlimited users, workflows, and concurrent executions, and you only pay when a numbered step runs. Triggers, waits, and approvals stay free, and the pricing page spells out how retries and loop iterations are counted.

Breyta says it's GDPR compliant and SOC 2 Type II compliant, with production workloads running primarily in EU regions on Google Cloud Platform. Access is scoped by workspace roles, and webhooks support HMAC, signature, bearer-token, and API-key authentication. It also states it doesn't train its own models on customer content.