Skip to content

Welcome

Delta AI is an architecture-first backend generator. You describe your product, answer a short round of questions, and it designs a backend system, writes the specification for every service, and opens the implementation as pull requests in your own GitHub organisation.

This page tells you what the product produces and how the rest of the documentation is organised.

A project produces a complete backend system, not a code snippet:

  • A system design — the services, their boundaries, their data ownership and how they talk to each other, with an overview diagram of the whole system.
  • A specification per service — architecture, an OpenAPI contract, testable user stories, an engineering guide, architecture decision records, and a diagram of that service’s internals.
  • An implementation per service — source, tests and a diagram, delivered as a pull request you review like any other.

Everything lands in your own GitHub organisation. You deploy the containers wherever you want. There is no runtime lock-in.

The agent does not guess and build. It works in stages, and you are in the loop at each one:

  1. You describe the product. A paragraph is enough to start.
  2. The agent asks what it still needs. Audience, scale, moderation, what success looks like — a short round of questions, answered in the chat.
  3. You review the draft. The agent restates what it understood before anything is designed.
  4. The architecture is designed. Which services exist, what each owns, how they communicate.
  5. You approve or steer the plan. If plan mode is on for that step, the run pauses and waits for you.
  6. Specification and implementation run per service. Each service gets its own plan, its own thread, and its own pull request.

How much of that a session touches depends on what you asked for. A feature may add a service and change two others; a refactor may touch one file in one service. See Sessions and chat modes.

The sidebar follows the order you actually work in:

  1. Getting started — sign in, install the GitHub App, understand the workspace, create your first project.
  2. Building a backend — the prompt, the clarifying questions, plan mode, provider research, run statuses, and what the agent does while it works.
  3. Reviewing the output — sessions, change sets, and how to review and approve what was generated.
  4. Account — credits, holds, billing and plans.
  5. Reference — a worked use case, glossary, FAQ, troubleshooting and the changelog.

The fastest way to understand the output is to read one. A full Discord-style platform was built and published openly, with its entire history intact: