An AI that writes code
Describe what the program should do and get working code in Python, JavaScript, HTML/CSS or SQL. Paste someone else’s code to have it explained, or a traceback to find the bug. A text scenario that works on the free plan.
Prompt sent to the chat: «Write a Python script that reads orders.csv with the columns client and amount and prints the total spent by each client, highest first»
The Starter plan is free: 30 units of energy, and a code request usually takes 1–2. After sign-up the chat opens with this prompt.
# Prompt: total purchases per client from a CSV
import csv
from collections import defaultdict
totals = defaultdict(float)
with open("orders.csv", encoding="utf-8") as f:
for row in csv.DictReader(f):
totals[row["client"]] += float(row["amount"])
for client, total in sorted(totals.items(), key=lambda x: -x[1]):
print(f"{client}: {total:.2f}")Not just write, but understand and fix
An AI for code is most useful where you need to quickly make sense of someone else’s code or broken code.
Explain someone else’s code
Paste a function from a project you inherited and write: “Explain line by line what this code does and where it can fail”. You get a plain-language walkthrough, a list of inputs and outputs and the suspicious places. Handy for reviews and for reading libraries with no docs.
Find the bug from a traceback
Copy the whole error together with the code that caused it. The AI reads the traceback, points to the failing line, explains the cause and suggests a fixed snippet. A screenshot of the console works too: upload the picture.
Write from scratch
A script to rename files, a table parser, an SQL query for a report. Describe the input, the result, the language and version. The code comes with comments and an example of how to run it.
An HTML website in one prompt
The AI builds the whole page: markup, styles and simple scripts in one file. Open it in a browser and the business card site, service landing or event page already works. Then edit in words: “make the button purple”, “add a form with name and phone”.
- Describe the structure block by block: the page comes out coherent rather than a set of random sections.
- Ask for “no external libraries” if you just want to open the file with a double click.
- For edits, name the block: “in the pricing section, replace the second item”.
For study and for work
For study
A Python lab, a sorting exercise, a query against a practice database. Do not ask “solve it for me” — ask “explain the solution step by step and give me a similar problem”. The AI shows why the loop works the way it does and reviews your attempt if you paste it. A useful prompt: “Find the bug in my code, but do not fix it — tell me where to look”.
For work
A marketer needs a script that merges ad-platform exports; an analyst needs SQL with a window function; a manager needs a spreadsheet formula. Describe the data structure (column names, a sample row) and the expected result. For recurring jobs, ask for the code as a function with parameters. You can also hand a colleague’s snippet over for a review against a checklist.
What to write so the code works the first time
- A file script
- Prompt: “Write a Python 3.12 script that renames every .jpg in a folder by its EXIF capture date”
- What to add: The language version and OS, so you do not get code for another environment
- An SQL report
- Prompt: “Table orders(id, client_id, amount, created_at). Show the top 10 clients by total over the last month, PostgreSQL”
- What to add: The database dialect and table schema
- A fix
- Prompt: “Here is the code and the traceback. Explain the cause and give a fixed version, marking the changes with comments”
- What to add: The full error text, not a summary
- An explanation
- Prompt: “Explain this code as you would to a junior: what it takes, what it returns, what side effects it has”
- What to add: The level to explain at
- Tests
- Prompt: “Write pytest tests for this function, including edge cases”
- What to add: The test framework you use
Check and test: AI code looks confident even when it is wrong
Run it before you rely on it
The AI writes code, it does not run it. It may refer to a method that does not exist in your library version. Run it, and if it fails, paste the error back into the chat.
Do not paste secrets
Replace passwords, tokens and API keys with placeholders like YOUR_TOKEN before sending code to the chat. It does not affect the result.
Big projects — piece by piece
The chat does not see the whole repository. Paste the specific function and its surroundings: imports, data structures, an example call. The answer is more precise and uses less energy.
A free AI for writing code
Code is a text scenario, so it is part of the free Starter plan. Paid plans are for writing code every day or generating images for your site in the same chat.
Starter
0 ₽, no card
30 units of energy. A code or explanation request takes 1–2 units: enough for 15–20 tasks — a few labs or a couple of work scripts.
All plansLite
₽690 a month
150 units of energy for daily work: scripts, reviews, SQL, plus image generation for your site from 1 unit. Basic at ₽990 and Premium at ₽2,290 for bigger volumes.
Get LiteFAQ
Is this really a free AI for code?
Yes. The Starter plan gives 30 units of energy with no card, and code is a text scenario. You pay only when the energy runs out or you need images.
Which languages are supported?
Python, JavaScript and TypeScript, HTML and CSS, SQL, PHP, Java, C#, Go, Bash, plus spreadsheet formulas and regular expressions. Quality is lower for rare languages, but the AI knows their syntax and logic.
Which model writes the code?
Several text models are available in the chat; you can pick one before the request. For complex tasks with a large context, choose a stronger model — it uses a little more energy.
Can I upload a code file instead of pasting text?
Yes, a file or a screenshot of the editor. For long files it is better to paste the relevant fragment: the model works with the context more precisely.
Can it build a multi-page HTML site?
Yes, ask for one page per request and a shared stylesheet. The AI keeps the header and menu the same if you remind it in each request or attach the first page.