Security and reliability

Your client data stays safe, and the system stays reliable.

From the day your AI delivery system goes live, it's secured with enterprise-level security architecture.

Illustrative Visual Studio Code project on a laptop showing redacted development secrets in a .env file, a .gitignore excluding local secrets, and a note that production uses a secrets manager.

Where home-made tools fail

5 things that go wrong with vibe-coded tools.

  1. Security keys and APIs are open to the public

    API keys and passwords pasted into the code, a shared doc or a Slack message. Anyone who finds them can get into your clients' accounts.

  2. Too much access

    Usually, all employees share one login, and it's impossible to track who made what changes.

  3. Client data sent to odd places

    Client files go to whichever AI service or plugin was easiest to connect, with no one checking what that service keeps.

  4. No backup and high failure

    Tools often change their login method, and if it fails, you are logged out. There is no copy to roll back to.

  5. Nobody is on call

    The person who made it has client work too. When it stops at 9 pm, there's no one whose job it is to fix it.

How we keep your clients' data safe.

  1. Hosting

    Your agency hosts it and owns it

    The system runs under your hosting and your own AI accounts, and you own everything, including the data.

    Illustrative Slack conversation: Vai reassures Alex that Alex's agency hosts and owns its system, controls access to client data, and can remove Vai's access at any time while keeping the system running.
  2. Keys

    Keys are locked away

    API keys and logins are kept in a secrets store, never in the code, a shared doc or a chat message.

    Illustrative secrets store showing an API key and service login with their values masked and kept in the secrets store.
  3. Logins

    A separate login for each team member

    We set up the system so every member of your team has their own login. You can easily track who made what change.

    Illustrative agency web app showing five named team accounts with access roles and twelve timestamped activity entries, including sign-ins, workflow changes, report approvals and access removal.
  4. Access

    Only the access it needs

    Each connection gets the smallest access that lets it do its job. If a step only needs to read a report, it can't delete anything.

    HubSpot, Facebook, Google Search Console and Google Analytics each supply a selected piece of data through a locked, read-only access gate to the AI delivery system. Other data remains in the source apps.
  5. AI models

    Your data doesn't train AI models

    We use the AI providers' business APIs, whose terms say the data you send isn't used to train their models.

    A laptop displaying an illustrative client data policy in Google Docs, explaining that client data sent through business APIs is not used to train AI models.
  6. Review

    A person checks before the client sees it

    The system hands the output to someone on your team. Your people check it and keep the judgement, so nothing reaches a client unseen.

    A team member reviews a draft client presentation in Microsoft PowerPoint, with slide thumbnails and a comment asking for review before sharing with the client.

Reliability

How we keep it running.

  1. Tested on real work before go-live

    Your team tries the system on test work like their real accounts, including the awkward clients, before anyone relies on it.

  2. Modular steps

    Each step is separate and tested, so a change in one tool doesn't quietly change the rest of the output.

  3. Monitored with New Relic

    We watch the system with New Relic application monitoring. When a login expires, a step fails or a run slows down, we get an alert. Your client doesn't become the error report.

  4. Backed up as often as you choose

    The system runs on your hosting, so you decide how often it's backed up, and it goes in your contract. We recommend every day. If something goes wrong, the system and its data can be restored.

  5. A reply within 5 minutes during hyper-care

    For 90 days after go-live, we reply within 5 minutes, 10:30 pm to 10:30 am EST, and we check the error logs every day, so most problems are fixed before your team notices.

    Fixes are free for 90 days, even when one of your tools changes.

  6. Every change is tested first

    Nothing changes until you agree, and we test every change before your team uses it.

Side by side

A vibe-coded tool vs our AI delivery system.

Both can use the same AI models. The difference is how the system around the model is secured and looked after.

A vibe-coded toolOur AI delivery system
API keys and loginsOften in the code or a shared docKept in a secrets store, never in the code
Access to client accountsWhatever login was quickest, often adminThe smallest access each step needs
Where client data goesWhichever service was easiest to connectUsually your agency's own hosting and AI accounts
AI training on your dataDepends on which app or plan was usedBusiness APIs whose terms exclude training on your data
Before go-liveTried on a few tidy examplesTested by your team on work like their real accounts
MonitoringNothing watching itNew Relic application monitoring, with alerts
BackupsUsually noneDaily backups recommended, set in your contract
When it breaksWhoever made it fixes it, if they're freeA reply within 5 minutes, 10:30 pm to 10:30 am EST, and free fixes for 90 days
When the person who made it leavesNobody knows how it worksWalkthrough videos, and the system stays with you

Ownership

Who owns what.

  1. The system is yours

    Usually your agency hosts and owns the system, as set in your contract.

  2. Your data stays yours

    Your clients' data stays in your tools and accounts. We only use it to run and fix your system.

  3. Running costs come straight to you

    Hosting and AI-model usage are billed to your own accounts by the providers. We don't add a margin.

  4. When our work ends

    After hyper-care, keep us on retainer, or run the system yourselves with the walkthrough videos. If you don't keep us on, you remove our access and the system keeps running for you.

In writing

No system is 100% secure. We'd rather say so.

Anyone who promises zero risk is selling something. Even trillion-dollar companies have seen security issues. What we can promise is how we set up, monitor and look after your system, and those details go into your contract: where it's hosted, who owns it, how often it's backed up and how fast we reply.

If your clients ask security questions we haven't answered here, send them to us before you sign. We'll answer in writing.

Questions

Questions agencies ask about security.

Where does our clients' data live?

Usually on your agency's own hosting and in your own tools and AI accounts. The exact setup is agreed in your contract.

Who owns the system?

Usually your agency does. Ownership is written into your contract, and you keep the system even if we miss the number in the guarantee.

Will our data be used to train AI models?

No. We use the AI providers' business APIs, and their terms say the data you send isn't used to train their models.

Who at EfficiaLabs can see our data?

Only the people working on your system, and only to set it up, test it and fix it.

How do you know when something breaks?

We monitor the system with New Relic application monitoring, so a failed step or a slow run sends us an alert. For the first 90 days we also check the error logs every day.

How fast do you reply?

During the 90 days of hyper-care, within 5 minutes, 10:30 pm to 10:30 am EST. Fixes in that time are free, even when one of your tools changes.

Do you back up the system?

Yes. How often is set in your contract, and we recommend a daily backup.

Can you work with our security policy?

Yes. Send it to us before the work starts, and we'll set the system up to fit it, or tell you in writing which parts we can't meet.

Next step

Have a tool you don't trust with client data? Show it to us.

Book a discovery call and walk us through the workflow, and the tool if you have one. We'll tell you honestly where the risks are and whether an AI delivery system is the right fix. If it isn't, you'll hear that too.

Talk to Vai (co-founder)

One call. If we go ahead, the Before-You-Hire Guarantee covers you.

Not ready to talk yet?Get 230 free AI skills

Talk to Vai (co-founder)

Before you go

Not ready to talk? Get your next client deliverable out faster anyway.

Take the 230 AI skills I give agencies for free. They work in Claude, ChatGPT and Gemini, in 10 folders, one for each service you sell.

Use your agency email. I'll send the link there.