# Rails: the jobs Kina does on repeat | BYOM

URL: https://byom.co/rails  
Markdown: https://byom.co/rails.md  
Last updated: 2026-10-03

> A Rail is a job Kina does for you on repeat, checked at every step and approved by you before anything touches your store. Every run ends with a record of what happened.

## What a Rail is

A Rail is a job Kina does for you on repeat, checked at every step and approved by you before anything touches your store. Every run ends with a record of what happened.

- On repeat: The same job, done the same careful way every time you start it.
- Checked at every step: BYOM tests what Kina produces at each stage, and a separate reviewer checks the parts only a careful reader would catch. If a check fails, the run stops there and tells you why.
- Approved by you: Anything that would change your store, your helpdesk or your campaigns waits for a named person to approve it.
- Ends with a record: What happened, who approved it, what your systems confirmed, and a way back where the system allows.

## One run, stage by stage: Where is my order?

A customer asks where their parcel is. Without a Rail, someone opens the helpdesk, finds the order in the store, checks the carrier, works out what is true, writes a reply and remembers to follow up. With a Rail, Kina does that job the same way every time, and BYOM checks each part before the next part can start.

- 1. Starting: The run is allowed to begin: the Rail is switched on, the person or trigger that started it is allowed to, and the tools it needs are connected. Checked before it can pass: The Rail is switched on and inside today's limits; Your store and helpdesk are connected; If something is missing, you see the one fix.
- 2. Gather: Kina pulls the customer, the order, the fulfilment and the tracking from the systems you already use. Checked before it can pass: Facts came from your connected systems; They are recent enough to trust; They belong to the same customer and agree with each other.
- 3. Prepare: Kina writes the reply, or works out the right next step. Checked before it can pass: The draft only says what the evidence supports; It follows your policies and tone; It stayed inside its budget.
- 4. Review: The reply arrives as one card in your Workspace, with the evidence beside it and the proposed action. Checked before it can pass: The evidence has not changed since the draft; The action is inside the rules you set.
- 5. Approve: You, or someone you have named, approve it. Or decline it. Or let it expire. Checked before it can pass: The right person decided; What was approved is exactly what will be sent.
- 6. Apply: BYOM sends the reply through your helpdesk, once. Checked before it can pass: It goes through the approval path; It is sent once, never twice; A stop switch is always in reach.
- 7. Proven: BYOM reads back from your helpdesk to confirm it really happened, then records it. Checked before it can pass: Your system confirmed it; Kina saying it is done is never enough.

## How BYOM checks the work

- Checks test the work, not the working: Kina is not a script, so BYOM does not check which steps it took. It checks what each stage produced: is the evidence complete and fresh, does the draft cite its sources, is the proposal inside your rules, did your system confirm the change.
- Done means your system says so: A run is only complete when the store, helpdesk or ad platform confirms the change. Commerce has more outcomes than pass and fail, so a Rail records unknown, and results that still need checking, instead of showing a green tick.
- A Rail cannot mark its own homework: Every Rail carries the checks it must pass. Only a person can change a check, and every change is recorded.
- Your approval is a decision, not quality control: The checks run before anything reaches you, so you are deciding on work that already passed. Nobody is asked to proofread a machine.
- Your decisions stick: Reject a proposal and the next run will not quietly propose it again. A control that the next run undoes is not a control.

## When a check fails, the run stops.

The run halts at the stage that failed, the trace shows which check failed and why, and you are told. Nothing after that stage runs, and nothing reaches your store.

## Rails compared with skills, automations and agent builders

Skills and Rails work together. A skill is how to do a task. A Rail is the whole job, done on repeat, with a check at every stage and your approval before anything changes. Kina can use skills inside a Rail; the Rail is what makes the result safe to rely on.

- What you get: An AI skill: Instructions an AI can follow; An automation: A fixed “if this, then that” script; An agent builder: A blank canvas to build an agent; A Rail: A finished job, run end to end
- Who does the thinking: An AI skill: The AI, once asked; An automation: The script, not a judgment; An agent builder: Whatever you build; A Rail: Kina, inside each stage
- Who checks the result: An AI skill: You, by reading it; An automation: Only what you script; An agent builder: You, if you build checks; A Rail: BYOM, before each stage can pass
- Before your store changes: An AI skill: Depends on the tool; An automation: Runs as scripted; An agent builder: Depends on what you built; A Rail: A named person signs off
- Proof it finished: An AI skill: The answer itself; An automation: A log line; An agent builder: Whatever you log; A Rail: Your system confirms it, then it is recorded
- Your decisions stick: An AI skill: Not tracked; An automation: Not tracked; An agent builder: If you build it; A Rail: Always: rejected work is not proposed again

## What stays in your control

- Start it when you choose: Run a Rail from its page, as often as the job needs, with a dry run first if you want one.
- Dry run first: See what a run would propose with the writing switched off. Nothing is sent and nothing changes.
- Stop at any moment: One button stops a run. If something was already sent, BYOM checks what actually happened rather than pretending it stopped.
- Limits you set: Daily run limits, spend caps and the systems each Rail may touch.
- Only what you connected: A Rail only reads the tools you connected, with the access you granted, and only changes them through approvals.
- A trace for every run: One row per stage with the checks that ran and whether they passed, like the status checks on a code change.

## The Rail library

- Where is my order? (the first Rail): Answers every question about order status with the real shipping status, and follows up when a parcel is stuck.
- Product data quality: Keeps every product complete and accurate: materials, sizing, care and specs, drafted from your own sources.
- AI shopping readiness: Keeps your catalogue easy for AI shoppers to read, and catches drift before it costs you visibility.
- Inventory risk: Spots stock running short or sitting too long, set against demand, returns and campaigns.
- Campaign vs stock: Checks paid campaigns against the stock and margin behind them before spend runs ahead of supply.
- Returns retention: Turns more returns into exchanges or credit, inside your returns policy.
- Support to action: Turns what customers keep asking into fixes on the product pages and policies that caused it.
- Readiness to sell abroad: Finds the currency, tax, shipping and listing gaps before you open a new market.
- Market expansion: Scores which new markets you are ready for, from your catalogue, channels, payments, policies and language.

## For technical teams

Kina does each stage, BYOM tests the output before it can pass, a person approves anything that writes, and every run ends with a record of what happened. Nothing reaches your store from a stage that has not passed.

- Versioned workflow → A Rail version with readable instructions
- Required status checks → Checks on what each stage produced
- Tests the author cannot edit → Checks the Rail cannot change
- Required reviewer → Your approval before anything writes
- Build log → The run trace, with stop
- Release artifact → The receipt

## Questions

- What is a Rail? A Rail is a job Kina does for you on repeat, checked at every step and approved by you before anything touches your store. Every run ends with a record of what happened.
- Does a Rail run on its own? It runs on repeat, never unattended. Kina does each stage, BYOM checks the output before it can pass, you approve anything that changes something outside BYOM, and every run ends with a record of what happened.
- How is a Rail different from an AI skill? A skill is instructions an AI can follow for one task. A Rail is the whole job: done on repeat, checked at every step, approved by you before anything changes, and finished only when your system confirms it. Kina can use skills inside a Rail.
- How does BYOM know the work was really done? Every stage produces something BYOM can check, and a run only completes when your store, helpdesk or ad platform confirms the change. Every run has a trace and every change has a receipt.
- What happens when a check fails? The run stops at that stage, the trace shows which check failed and why, and you are told. Nothing after that stage runs and nothing reaches your store.
- Can I stop a Rail? Yes, at any moment. If an action was already sent, BYOM checks what actually happened and records that.
- Can I build my own Rail? You start from a Rail that comes built and checked, and Kina helps you set it up for your store. Your own Rails come later, because the checks are the product.
- Which Rails can I run? Where is my order? is the first Rail, and it runs end to end. Each Rail after it launches with its own checks and a completed real run behind it.
- What does the judgment engine do in a Rail? It checks what each stage produced: that a draft only says what the evidence supports, follows your policy and avoids what you told Kina never to say. It never approves a change. That stays with you.
- What does a Rail cost? Rails are part of your BYOM plan. Kina's model usage counts against your credits, or runs on your own model keys.
- Does a Rail use one of my scheduled tasks? No. A Rail is a job you start when you choose, so it does not take a scheduled task. Scheduled tasks are the jobs Kina runs for you on repeat, such as a weekly check on your board. Explore has 5, Starter 15, Growth 25 and Scale has no limit.
