Skip to content

Create your first project

This page takes you from an empty workspace to an approved architecture. It is the shortest path to seeing what Delta AI actually produces.

You need a signed-in account and the GitHub App installed on your organisation — see GitHub App installation. The first architecture and one service end to end are free, so you can complete this page without adding a card.

Type what you want to build into the composer. Plain language is correct here — you are describing a product, not writing a specification.

Composer with a product description typed in, ready to send
The composer with a product description. One paragraph is enough; the agent asks for anything it still needs.

A good description names the users, the core actions and the constraints that matter. A bad one names a technology or a framework — those are decisions the agent will research and propose.

Before the agent starts, you pick the implementation language. This applies to every service in the project.

Programming language picker showing TypeScript, Python, Go and Rust
The language picker. The choice applies to every service generated in this project.

The agent asks what it needs to know before designing anything: audience, scale, moderation requirements, what success looks like. Answer them as you would to a colleague.

Clarifying questions card showing a question about the primary audience with selectable answers
Clarifying questions. Each answer narrows the architecture, so the design does not have to guess.

You can select a suggested answer or write your own. The agent will not proceed until the questions are answered.

Once the questions are answered, the agent produces a draft of what it understood. This is your last chance to correct the brief before the workflow locks it.

Draft review panel showing the interpreted requirements with a submit action
The draft review. Correct anything the agent misread before the brief is locked.

The workflow starts. The first step designs the system: which services exist, what each owns, and how they communicate.

Live activity feed showing the architecture agent reasoning and calling tools
Live activity during architecture. Every reasoning step and tool call is visible as it happens.

If plan mode is on for the architecture step, the run pauses and presents the plan for approval. You can approve it, or steer it with a message.

Architecture plan approval card showing the proposed services and per-service credit estimates
The architecture plan, with the proposed services and their estimated credit cost.

When the architecture completes, the system design is available in the State tab alongside a diagram of the whole system.

State tab showing the system design document next to the architecture diagram
The system design and its diagram, side by side in the State tab.

From here the workflow continues into specification and implementation. See How to build a backend for the rest of the journey.