Skip to content
AI & models

What Codexe does behind the scenes

The steps between your request and a Ready project: continuing cut-off replies, the no-output retry, platform checks, repair with the real API, and what you pay for.

Updated

On this page
  1. 1. The generation
  2. 2. Continuation (free)
  3. 3. The no-output retry (free)
  4. 4. Platform check (free)
  5. 5. Build or load test (free)
  6. 6. Repair with the real API (free)
  7. What you pay for

A lot happens between pressing Generate and a Ready project. Most of it is work Codexe does at its own cost so you get a complete, building project.

#1. The generation

The model writes the project in one reply, starting with a plan of the files it's going to write. You're charged for this reply.

#2. Continuation (free)

Models sometimes stop mid-file, or skip a file they planned. Codexe detects this (an unfinished code block, a missing planned file) and asks the same model to continue, up to four times, until every planned file exists. You'll see Writing remaining files…. Continuation isn't billed.

#3. The no-output retry (free)

If the first reply describes the project instead of writing files, Codexe asks once more, explicitly, for the files. The same pass also cleans up placeholders such as // TODO. Not billed.

#4. Platform check (free)

Before building, Codexe checks the files are for the platform you asked for. A Fabric mod delivered for a Spigot request would compile, but it isn't what you wanted. If the platform is wrong, Codexe regenerates it once with a stronger model. If that still fails, the project stops and, on a first generation, your tokens are restored.

#5. Build or load test (free)

Plugins and mods are compiled; datapacks and Skript are loaded on a real test server. See Compile.

#6. Repair with the real API (free)

When the build fails, the repair model sees the exact errors and the real public signatures of the library classes involved, read from the same JARs the build used. That's how it fixes "method not found" errors for the API version you're actually on, rather than guessing. See Auto Repair.

#What you pay for

StepWho pays
The generation replyYou (tokens)
Web searches the model actually completes (Business+)You (tokens)
Continuation, no-output retry, platform regenerationCodexe
Builds, load tests, repairs, the escalation to a stronger modelCodexe
Extra output on a Free first generation above the 4,000-token capCodexe
A first generation that delivers nothing usableRestored to you. See Credit restoration
What Codexe does behind the scenes | Codexe Docs