I run thirty side projects. Most coding-agent products are designed for someone running one.
That's the gap.
The shape of the problem
When you have one repo, the question is: how do I write more code per hour? Inline-completion editors, AI pair-programmers, agentic IDEs — those all answer that question. They make you faster while you sit at the keyboard.
When you have ten repos, the question changes. It becomes: how do I keep each one alive without sitting at any keyboard for very long? The constraint isn't speed-per-keystroke. The constraint is continuity of momentum across products.
You're walking between meetings. You're at a coffee shop without your laptop. You're putting your kid to bed. Three different products each have a small bug or a small feature that would take an hour at a desk. None of them are getting that hour today.
What a real afternoon looks like
You notice a typo in your checkout flow. You open FlareCode on your phone, type "fix the typo in the success message on the payment page," and submit.
The agent boots an isolated sandbox, clones the repo, finds the file, edits it, runs the tests, and opens a Pull Request. You get a notification. Ten minutes later you're in an elevator, review the diff on your phone, and merge.
The fix ships before you reach your desk.
The full loop
- 01
Submit
Describe the task in plain English from any device — no template, no DSL.
- 02
Isolate
The agent boots a fresh sandbox, clones your repo, and starts working.
- 03
Edit & test
It writes code, runs your test suite, and fixes failures automatically.
- 04
Open PR
A branch named flarecode/task-<id> is pushed and a Pull Request is opened.
- 05
Review
You review the diff on your phone and merge — or close it if it's wrong.
What FlareCode does instead
You describe a task in plain English. The agent handles the rest in an isolated sandbox — clone, edit, test, PR. You get a notification when it's done. You review the PR from your phone, your laptop, or wherever you are.
That's it. No terminal session. No babysitting. No "is the agent stuck?" loop. The agent finishes, or it fails and tells you why.
~10m
avg task time
0
installs on your machine
1
PR per task
25
projects on Fleet
What it deliberately doesn't do
- It doesn't push to your default branch. Branch protection stays load-bearing.
- It doesn't auto-merge by default. Pro plan customers can opt in per-repo.
- It doesn't treat your code as training data. Agent workspaces persist as private encrypted snapshots so work can resume, and are deleted when you destroy the agent.
- It doesn't pretend to know your stack better than you do. If the PR is wrong, close it.
- It doesn't charge you for time the agent spends idle. You pay for the task, not the sandbox uptime.
We never touch your default branch. Every change lands on a feature branch. Your existing branch protection rules, required checks, and review policies remain fully in force.
Who it's for
Builders who don't sit still. People running multiple products, shipping between meetings, reviewing code from a phone. If your ideal workflow is "hand off the task and keep moving," this is built for you.
The product is not for the developer who wants to type faster. It's for the builder who wants to keep shipping while they're walking.
The bet
Hosted runtime + multi-repo fleet + flat-rate inference is the wedge. Each piece on its own already exists somewhere. The combination, aimed at builders who don't sit still, doesn't.
We'll see if that's true. The open beta is live.