Creating a project

Every project starts on Create. This page walks through each part of it.

#Project type

The type picker groups every supported target:

The type controls the project scaffold, the build tool and how Codexe verifies the result. If your request clearly describes a different type ("a Fabric mod that…" while Spigot is selected), Codexe switches the type and shows a notice so you can switch back.

You can also change the type later, from the chat: ask for it ("make it a Forge mod", "turn it into a Skript", "I want a website instead"). Codexe converts the whole project: between plugins and mods it ports the code, between different kinds (plugin or mod, Skript, website) it rewrites the project for the new type with the model you picked and charges the tokens that rewrite uses. The new type gets its own check: compile for plugins and mods, a test-server load for Skript, none for websites.

#The request box

Write what you want. How to write a good request covers what to include. There's no separate Minecraft-version selector: write the version in the request ("Paper 1.21.4", "Fabric 1.20.1"), and Codexe picks the matching Java version for the build.

#Model and reasoning effort

The default model is chosen for your plan. Paid plans can pick other models and a reasoning effort (how much the model thinks before answering; higher effort costs more). Free accounts use the default effort. See Models.

#Attachments

Use +, paste or drag and drop to attach images (a screenshot of an error, a mock-up of a menu, a texture you want matched) and, on paid plans, text and code files (existing classes, config files, logs). The model reads them together with your request.

• Images are scaled down in your browser before they are sent (longest side 1568 px): the model sees the same picture, and it costs fewer tokens.
• Files must be text: code, configs, logs, Markdown. Archives, PDFs and other binaries cannot be attached; to start from a whole project, use Importing a project.
• Attachments cost tokens: what the model reads for them, plus 3,000 per image and 1,000 per file. See Tokens.
• Going past your plan's limit shows what the next plan allows before anything is sent or charged.
• Codexe keeps your attachments with the message: click an image or a file in the chat to open it again. They are deleted with the project or the account. Attachments sent before 2 October 2026 show only their name.

To start from an existing project, see Importing a project.

#Questions before Codexe starts

When your request leaves open a choice that really changes the result (for example two very different ways a feature could work, or the art style of a model), Codexe asks before generating: one to three short questions, each with a few answers and one marked Recommended. Pick an answer or write your own under Other. Your answers are added to the request, so you can see them in the chat.

• Skip, let Codexe decide generates with Codexe's own choices; Back returns to the request box.
• Most requests get no question and start at once.
• Asking is free: no tokens are charged for the questions.
• The same can happen in the project chat: when a follow-up is ambiguous, the AI answers with questions instead of changing files. That reply costs the tokens the model used, like any other answer; your choice is sent as your next message.

#Enhance

Enhance rewrites your request into a more complete specification using a fast model. It costs a flat 5,000 tokens. Read the result before generating: it may add assumptions.

#What it costs

A generation costs what the model actually reads and writes, settled when it finishes; the chat shows the cost of each message. See Credits.

#One generation at a time

Each account runs one generation at a time. If another of your projects is still generating, wait for it to finish. There are also fair-use limits on how many generations you can start per hour (30 per hour, fewer on the most expensive models and highest reasoning efforts).