GPT for developers

A quick second pair of eyes on the code in front of you: unfamiliar modules explained, bugs narrowed down, layouts fixed from a screenshot and tests you forgot to write.

4 min readUpdated

GPT.me works as a quick second pair of eyes for code: paste a function, an error or a diff and get an explanation, a hypothesis or a test within seconds. It is free and needs no sign-up. GPT-6 Luna is fast enough for everyday questions. GPT-6.1 Terra thinks briefly before it answers, which helps with logic bugs and design trade-offs, and it is good at reading screenshots. Answers come as Markdown with highlighted code blocks. Never paste secrets: your messages pass through our server to the model provider. Here are seven uses with prompts to open in the chat.

Understand and fix code

Four uses for the code you are looking at right now, from a first read to the tests that keep it working.

Get a guided tour of unfamiliar code

Inherited a module with no documentation? Ask for a reading order and a map of how data moves through it before you change anything.

I'm new to this [language] code and need to change [part]. Explain what it does at a high level, then trace how data flows from [entry point] to [output]. List the functions in the order I should read them, and point out anything surprising, such as hidden side effects or global state: [paste code]

Start the tour. The Code Assistant is handy for one-off snippets.

Narrow down a bug from its symptoms

When a bug shows up only sometimes, ranked hypotheses beat guessing. GPT-6.1 Terra reasons through the evidence before it answers.

There is a bug in my [language and framework] app: [describe the symptom and when it happens]. Here are the relevant logs and code: [paste logs and code]. Give me up to four hypotheses ranked by likelihood. For each one, say what evidence supports it and the cheapest experiment that would confirm or rule it out.

Debug with GPT-6.1 Terra, run the first experiment and report back in the same chat.

Fix a layout from a screenshot

Describing a broken layout in words is hard; showing it is easy. Attach a screenshot together with the markup for that part of the page.

I've attached a screenshot of a layout problem: [what is wrong and what it should look like]. The markup and styles involved are: [paste HTML and CSS]. Tell me what causes the problem and give me the smallest CSS change that fixes it without breaking other screen sizes.

Open it with GPT-6.1 Terra and attach the screenshot.

Write tests for the cases you forgot

The happy path is usually covered. Empty inputs, limits and odd encodings often aren't.

Write unit tests in [test framework] for this [language] function: [paste function]. Start with a short list of the edge cases you plan to cover, such as empty input, boundaries, invalid types and large values. Then write the tests, one behaviour per test, with names that describe the expected result.

Write the tests, run them, and paste any failures back into the same chat, which still has your function in context.

Design, data and housekeeping

Three uses for the work around the code: decisions, queries and the history you leave behind. None of them needs a long paste, so they fit comfortably within a single message.

Weigh two designs before you build

A short written comparison often settles a design debate faster than another meeting.

I need to [goal] in a [stack] project and I'm choosing between [option A] and [option B]. Compare them for complexity, performance, failure modes and how hard each is to change later, given these constraints: [constraints]. End with a recommendation and the one assumption that would change it.

Compare them with GPT-6.1 Terra and share the answer with your team as a starting point.

Write a SQL query and check it

Give the schema and a few rows, so the model can show the result it expects and you can spot a wrong join.

Here is my [database] schema: [paste CREATE TABLE statements]. Write a query that returns [what you need]. Explain each join and filter in one line, then show the rows the query would return for this sample data, so I can check the logic: [paste sample rows]

Build the query, then run it on a copy of the data first.

Turn a diff into a clear commit message

Good history helps the next person who reads it, who may well be you. Paste the diff and get a message in your team's style.

Write a commit message for this diff in the [convention] style: a summary line under 72 characters, a blank line, then a short body that explains why the change was made, not only what changed. If the diff mixes unrelated changes, tell me how to split it: [paste diff]

Write the message and trim it to what a reviewer needs.

What to avoid

  • Pasting secrets. Remove API keys, passwords, tokens and customer data before you send anything.
  • Trusting APIs from memory. The models have a training cut-off and cannot open current documentation, so a function may have been renamed or removed. Run the code and check the docs.
  • Sending the whole repository. Very long messages are trimmed. Send the function and the error that matter.
  • Merging without review. Treat every answer as a colleague's suggestion and read each line.

Got a question? Ask it now

No account and no download. Open the chat, type what is on your mind, and the answer starts appearing in a second or two.

Start chatting