SystemsBBBLYAI coding agent

Bubbly. Agentic engineering, with the proof attached.

AI-assisted engineering with the rigor left in: every change planned, tested and reported.

Talk to us about Bubbly How it works

How it works · BBBLY

A task's journey. From plan to proof.

01 / 04Read & planIt reads the code before it changes it, and says what it plans to do.
02 / 04Edit & runEdits land in the workspace you chose: a local folder, an SSH host or a cloned repository, with the model you chose.
03 / 04VerifyThe bug is reproduced before the fix and passing after it; the tests are run and their output read.
04 / 04Review the resultThe report keeps what passed apart from what was not checked. It can open a pull request; it cannot merge one.

Local folderSSH hostRepositoryModel: your choice

Workspace

src
parse.ts
routes.ts
tests
parse.test.ts
package.json

parse.ts

1export function parseRange(input: string) {

2 const [start, end] = input.split("-").map(Number);

3 if (Number.isNaN(start) || Number.isNaN(end)) return null;

4 return { start, end: end - 1 };

4 return { start, end };

7export const span = (r: Range) => r.end - r.start + 1;

›/fix parseRange drops the last value

Plan

Reproduce it with a failing test
Change parseRange
Run the tests, read the output

Terminal

$ npm test -- parse

parseRange › keeps the last value before the fix · reproduced

$ npm test

parseRange › keeps the last value after the fix · passing

Report/fix

Verified

Reproduced before the fix
Passing after it
Tests run, output read

Not checked

Repository map incomplete, said so

Pull request opened

CommentMerge

It can open one. It cannot merge or close one.

Verified · reported

What it is

Done means the evidence says so.

For engineering teams that use AI to write code and still hold it to their standards. Bubbly's workflows plan before they edit, reproduce a bug before fixing it, run the tests before a change is called done, and report what was verified and what was not. It works in a desktop IDE and the terminal, with the model you choose, a local one included, and it can open pull requests but never merge them.

An AI that writes code quickly still needs the habits of a careful engineer: read before changing, reproduce before fixing, test before claiming. Bubbly's workflows ask for those steps, and its report keeps what passed apart from what was not checked.

Sourceregnant-io/bubbly

AI-written code held to your standards?

Say what it should do, which data it can touch and where it has to run. A person replies by email.

Talk to us about Bubbly All systems