From ticket to release. Verified.

Loopback is the AI support engineer that reads your customer’s request, finds the exact place in your code and logs, writes the reply and prepares the fix. You approve; it releases and tells the customer.

Try it on your repository Read the docs

14 days free, nothing charged until the trial ends. Always free for public repositories.

Product

The ticket view your team approves from

The customer’s message, the cause in your code and the verified fix, on one page.

Inboxtkt_01k6h2qv8m3n7p9r4s5t6w8x0yTallybird

Error 500 at checkout

From Giulia Ferrarovia EmailOctober 1, 2026 at 9:47 AM

Awaiting approvalBugHigh

Approve and release

Your approval records your name, the time and commit 9c1e5b7. Merge, tag and release follow.

Pull request: #412
Commit: 9c1e5b7
Approvals: 0 of 1
Merge: Squash and merge when the required checks pass
Release: Tag v1.8.3 on the merge commit

Conversation

2 messages

Customer · Giulia Ferraro <giulia.ferraro@example.com>October 1, 2026 at 9:47 AM
Error 500 at checkout
This morning I tried to upgrade to the Pro plan, but when I press “Pay now” the page loads for a few seconds and then shows “Error 500”. I entered the welcome discount code from your newsletter. I need Pro by Friday to send my invoices.
Sent by Ada RicciOctober 1, 2026 at 9:52 AMDelivered
Hi Giulia, thank you for the details. We found the cause: percentage discount codes on monthly plans produced an amount the card processor refuses. The fix is being verified now, and we will write to you here with the version that contains it.

Diagnosis

Bug90% confidence

createPayment (src/billing/checkout.ts:46) subtracts the percentage discount without rounding to whole cents:

amountCents -= (amountCents * coupon.percentOff) / 100;

With BENVENUTO15 on the Pro plan (1999 cents) the amount becomes 1699.15 cents. validateChargeRequest accepts whole cents only and rejects the charge with amount_invalid, so POST /api/checkout answers 500. Every percentage coupon on a monthly plan is affected; fixed-amount coupons are not.

Fix: round the discount with percentOf from src/lib/money.ts and add a regression test.

Code

src/billing/checkout.ts:40-59createPayment
src/billing/payment-provider.ts:153-161validateChargeRequest

Links point to commit 4f0d2c9.

Known solution, seen in 37 apps with this stack

Round percentage discounts to whole cents before the amount reaches the payment provider.

node@22

Worked in 92% of the tickets that used it.

Proposed status page summary

Payments failed when a percentage discount code was entered. We found the cause and the fix is being prepared.

Fix attempts

1 attempt

Attempt 1

Waiting for approval

Branch: loopback/tkt_01k6h2qv8m3n7p9r4s5t6w8x0y-checkout-500
Pull request: #412
Commit: 9c1e5b7
Checked by: The runner harness
Changes: 2 files, +73 −1

Automatic verification

Passed

Test to runtest/checkout.test.ts:33createPaymentCovers the changed function

Tests

Passednpm test

# tests 58# suites 20# pass 58# fail 0

Change impact

createPayment has one caller, handleCheckout (POST /api/checkout). No other module uses the changed line.

createPaymenthandleCheckout

Diff

The diff is long: the first part is shown. The full diff is on GitHub.

src/billing/checkout.ts

+2−1

@@ -4,6 +4,7 @@ import type { Charge, ChargeRequest, PaymentProvider } from './payment-provider.
4	4	import { getPlan } from './plans.ts';
5	5	import type { Plan, PlanId } from './plans.ts';
6	6	import type { MemoryStore, Payment, PaymentDetails } from '../data/store.ts';
	7	Added: import { percentOf } from '../lib/money.ts';
7	8
8	9	export interface CheckoutInput {
9	10	email: string;
@@ -43,7 +44,7 @@ export async function createPayment(input: CheckoutInput, deps: CheckoutDeps): P
43	44
44	45	let amountCents = plan.priceCents;
45	46	if (coupon?.kind === 'percent') {
46		Removed: amountCents -= (amountCents * coupon.percentOff) / 100;
	47	Added: amountCents -= percentOf(amountCents, coupon.percentOff);
47	48	} else if (coupon?.kind === 'amount') {
48	49	amountCents = Math.max(0, amountCents - coupon.amountOffCents);
49	50	}

State history

Every state the ticket went through, from the audit trail.

ReceivedOctober 1, 2026 at 9:47 AM · Customer
Triage and diagnosisOctober 1, 2026 at 9:47 AM · Pipeline
Fix and verificationOctober 1, 2026 at 9:49 AM · Pipeline
Awaiting approval (current step)October 1, 2026 at 10:05 AM · Pipeline

Real product view with demo data: a payment error reported by email, diagnosed in the code, answered, fixed on a branch and verified. One click releases it.

See the six steps

How it works

Six steps. One decision. One approval.

Usage questions get their answer right away. Bugs go through a verified fix and your approval, and the customer gets the reply with the version that fixes them.

Received
Email, widget, GitHub issue or an error that hits many users: one ticket, and one status page for the customer.
Diagnosed
Function, file and line in your code and logs, with a confidence score. Low confidence goes to a person, not to a guess.
Decided
A usage, configuration or documentation question gets a reply. A bug gets a fix. This one decision is automatic.
Fixed and verified
On a loopback/ branch, never your default branch. Mechanical verification and your tests must pass, or no pull request is opened.
The only human step
Approved
One click from a person, recorded with user, time and commit SHA. Nothing is tagged without it.
Released and announced
Merge, tag and release on your pipeline, then the email to the customer in the same thread, with the version.

Verification

A person approves mechanical proof

Before a fix reaches you, Loopback checks it: files that no longer parse, imports that broke, code that depends on what changed. Then it runs your tests. The report lists files and lines, so the approval rests on facts, not on a model’s opinion.

When verification or tests fail, no pull request is opened: the ticket goes to your team with the report.

Verification report of the fix

Automatic verification passed

Parse errors: 0
Broken imports: 0
Dependents checked: 2
Tests: 58 of 58 passed

src/billing/checkout.ts:40-59
src/billing/payment-provider.ts:153-161

Engine

Powered by Vexp

The analysis runs on vexp, the local-first code-context engine for AI coding agents. It keeps a live dependency graph of your repository, so the agent reads the exact files and lines that matter: fewer tokens, better diagnoses. Every change gets its blast radius, and verify_done checks the fix mechanically before any pull request.

fewer context tokens in the public example: 8,247 become 2,140: −74%
on SWE-bench Pro with Opus 5, reference result: 81.7%
per task on SWE-bench Pro with Opus 5: $1.52

vexp.dev

Execution

Your code runs where you choose

Three execution modes. Approval, tags and the message to the customer always go through Loopback.

Managed
Where the code runs
A private container in our cloud, created for each job and deleted at the end
Who pays the model
Loopback, with tokens included in the plan
For
Teams that want zero setup
Always on
Yes
Your runner
Where the code runs
Your GitHub Actions or your self-hosted container
Who pays the model
You: your API key, your cloud provider, or a ChatGPT plan for Codex
For
Developers and teams that keep code in-house
Always on
Yes
In your agent (MCP)
Where the code runs
Your machine, inside your IDE or CLI agent
Who pays the model
Your agent’s subscription: Claude Code, Cursor, Copilot and others
For
Developers who already pay for an agent
Always on
While you work

Claude subscriptions run only in MCP mode, inside the official client, as the provider’s terms require.

Status page

Your customer follows the ticket to the release

Every reply links to a public page that updates by itself: under review, cause found, fix being released, resolved in the version. It never shows code, file names or internal details.

At the bottom of pages and emails, a small badge with the time it took. It stays on the entry plans, can be removed on higher plans and never appears on Agency.

Example status page

Around the fix

Everything around the fix

Reports that arrive with their context, errors caught before anyone writes, a memory shared with consent, and Loopback free on public repositories.

One script tag, or the SDKs for React Native, iOS and Android. Each report brings console errors, failed requests, browser and page; passwords, emails and card fields are masked in the browser before anything is sent.

under 15 KB gzip · no dependencies

index.html

<script src="https://cdn.loopback.so/w.js" data-key="pk_live_..." data-release="2.4.0" async></script>

Install the widget

Fixed before anyone writes

Loopback watches real errors from the widget, Sentry, Datadog or CloudWatch. When a group hits 5 users in 24 hours, it opens a ticket and prepares the fix, and after the release the affected users are told.

up to 10 proactive fixes a month per workspace

TypeError: Cannot read properties of undefined (reading 'id')src/invoices/export-pdf.ts:88

23 / 24h2.4.0/checkout

Shared memory, with consent

With your explicit consent, Loopback learns abstract patterns of errors and fixes on common stacks: never your code, customer names or personal data. Withdrawing consent removes your contributions.

shared only once seen in 3 different organizations

Free for open source

On public repositories Loopback answers GitHub issues with a diagnosis at file and line, finds duplicates and prepares verified fixes for maintainers to approve.

$0 a month on public repositories

lib/parser/tokenize.ts:214-231

Duplicate of #1874. Proposed label: bug.

Diagnosed by Loopback

How it works for maintainers

Agency

For software houses: your brand, your margin

One workspace per client, one approval queue, pages and emails with your brand, a monthly report per client and resale with your margin. Your clients never see Loopback.

a month for the plan: $299
client workspaces included: 10
a month for each additional client workspace: $19
approval queue for every client: 1

The Agency plan

Plans

Pricing

Bring your own agent or key at a flat price, or let Loopback run everything with tokens included.

Indie
For solo developers and small SaaS
$49/month
Tickets a month
150
Repositories
1
Autonomy
Copilot, Approve
Startup
For product teams
$199/month
Tickets a month
1,000
Repositories
5
Autonomy
Copilot, Approve, Autopilot
Growth
For scale-ups with many repositories
from $499/month
Tickets a month
Unlimited
Repositories
Unlimited
Autonomy
Copilot, Approve, Autopilot
Managed
Prefer we run it? Tokens included
$149/month
Fixes a month
25
Extra fix
$5
Tickets a month
100

Open Source is free for public repositories. Agency and Enterprise are on the pricing page.

Compare plans

FAQ

Questions before you connect a repository

Something else? Write to us.

Does my code leave my infrastructure?

With the runner or MCP, jobs run in your GitHub Actions, your container or your IDE, not in our cloud. Loopback receives the diagnosis, the reply draft and the verification report; diagnoses quote short code excerpts unless .loopback.yml sets share: references-only. One exception: the demo ticket after you connect a repository runs once in our sandbox, at our expense. With Managed, the code is cloned into a private container that is deleted at the end of each job.

What do you keep, and for how long?

Error events for 30 days. Screenshots and technical details of widget reports until 30 days after the ticket is closed. Prompts and outputs of jobs, already masked, for 30 days by default, configurable. Clones are deleted at the end of each job. Secrets are encrypted and only their last 4 characters are ever shown.

Do you train models on my code?

No. Your code is not used to train models, and Loopback uses the zero-retention options of the model providers where they exist.

What powers the analysis?

vexp, a code-context engine. It reads only the code that matters for the ticket, measures what a change affects, and checks every fix mechanically before it reaches you. The developer documentation lists its public benchmarks.

Can the agent push to my default branch?

No. Fixes go only to loopback/ branches, and a tag needs a person’s approval of the exact commit SHA. With Managed and the runner the agent holds no push credential: the push happens afterwards, with a short-lived token limited to that branch. In MCP mode your agent pushes with your own git credentials, and Loopback opens pull requests only from loopback/ branches.

What if the diagnosis is wrong?

Low confidence goes to a person instead of a fix attempt. A fix that fails verification or tests never becomes a pull request. Autopilot answers only the categories you choose, and never releases.

Which agents work with it?

In your runner: Claude Code, Codex, opencode, and the built-in agent. In Managed: Claude Code and the built-in agent. In MCP mode, inside your IDE or CLI: Claude Code, Codex, opencode, Cursor, Windsurf, Zed, Kiro, Antigravity, Continue, Cline, and GitHub Copilot.

How soon do I see it working?

After you connect a repository, Loopback runs a demo ticket on your code at its own expense, in its sandbox: if your workspace does not run there already, it asks once for your consent to clone the code. The first diagnosis arrives in less than 10 minutes.

Connect a repository. Approve the first fix.

The setup opens one pull request with the configuration: you merge it, and you never write YAML.

Try it on your repository

14 days free, nothing charged until the trial ends. Always free for public repositories.

From ticket to release. Verified.

The ticket view your team approves from

Six steps. One decision. One approval.

Received

Diagnosed

Decided

Fixed and verified

Approved

Released and announced

A person approves mechanical proof

Powered by Vexp

Your code runs where you choose

Managed

Your runner

In your agent (MCP)

Your customer follows the ticket to the release

Everything around the fix

A report button that brings the stack trace

Fixed before anyone writes

Shared memory, with consent

Free for open source

For software houses: your brand, your margin

Pricing

Indie

Startup

Growth

Managed

Questions before you connect a repository

Connect a repository. Approve the first fix.

From ticket to release. Verified.

The ticket view your team approves from

Six steps. One decision. One approval.

Received

Diagnosed

Decided

Fixed and verified

Approved

Released and announced

A person approves mechanical proof

Powered by Vexp

Your code runs where you choose

Managed

Your runner

In your agent (MCP)

Your customer follows the ticket to the release

Everything around the fix

A report button that brings the stack trace

Fixed before anyone writes

Shared memory, with consent

Free for open source

For software houses: your brand, your margin

Pricing

Indie

Startup

Growth

Managed

Questions before you connect a repository

Connect a repository. Approve the first fix.