Cascade roadmap

No license check exists anywhere in the code. Each phase ends when its “real” list has been done against real systems, not when the code compiles.

0.1.0, the MVP

Built. The engine, durable executions, the Postgres queue and workers, expressions, the Code step in its jail, the credential vault and OAuth2, webhooks, schedules and MQTT, users with enforced MFA, OIDC, projects and RBAC, audit, version history, the API, the editor, the connectors on the approved list, workflow import, metrics and traces, the Helm chart and the Quadlet units. Cut by RELEASE_CHECKLIST.md, not before.

0.2.0, identity and governance

SAML, LDAP and AD, SCIM, custom roles, external secrets, log streaming, variables, folders, insights, execution annotations, the S3 and Azure blob backends, and execution redaction policies. Each one is tried against the real server it talks to.

0.3.0, environments and scale

Git source control and environments, protected instances, the multi-main HA scheduler with trigger leases, the webhook role, concurrency and rate limit controls, retention, backup and restore. It is done when I can kill the leader main during a minute-level schedule and find no duplicate and no missed runs over an hour.

0.4.0, connectors and people

The rest of the approved connector list, the published declarative connector spec, a plugin SDK and index, and forms with human approval steps. Approvals have to work from a phone, because the person approving a recipe change is standing at the line, not sitting at a desk.

0.5.0, AI and MCP

LLM and agent steps, an MCP client and server, and a builder that uses your own model. Bring your own key or run it local. There is no hosted service and there won’t be one.

1.0

Stability promises on the workflow file format and the API, an upgrade path tested from every 0.x, and the docs.

After 1.0, a new editor UI

The editor is a hand-built TypeScript app, and it has done its job. The last phase on this list, done after everything above and not before, is rewriting it in Astro, the same framework the HotLoop website runs on, built to static assets and embedded in the Go binary the way the editor is now. It is the same one-binary story with nothing new to run.

The point is that every HotLoop project ends up with a modern UI that looks like it belongs to the same family, instead of each one carrying its own homegrown front end. A design pass comes first, with the layout, the components and the phone behavior settled on paper before a single page is built, because a rewrite without a design is just a more expensive way to get the same UI. Nothing about it changes the API, the file format or the Go side.

This one waits its turn. Nothing starts on it until the rest of the list is shipped.