Security and support: what IT teams ask us
Software built by AI agents should be held to the same standard as any other software your company runs. Here is where yours runs, what the agents can and can't do, how changes reach your team, and who you call when something breaks. Where we don't have something yet, we say so.
Sample chat
Where your software runs, and who owns it
You own the code and the data
Any developer can read it, change it, or move it.
It runs on a server you choose
Your own (no hosting fee to us), or one we run for you (quoted).
It runs on K2
K2 is the software we make for running AI agent teams. It is free to download and Fair Source licensed, so on your own server your software and its agents keep running even if you stop working with us.
Your AI plan is yours
The agents use the Claude, ChatGPT, Gemini or Grok plan you choose, billed by the AI company with no markup from us.
What the agents can and can't do
They work inside rules you set
For each kind of change, you decide: it goes live straight away, it waits for your approval, or it's tried on a test copy first.
Only a person can approve new access for an app
On current K2 servers running Linux (K2 version 0.46.1 or later), each app runs walled off. It reaches only what its owner approved. New access, new internet hosts and wider settings all go into one approvals list for a person. An agent can never approve its own request.
Every call an app makes is logged
The log shows who, what, and the answer, and it can be checked to prove no line was changed or removed. It covers what the apps do, not every action an agent takes.
Secrets stay in a vault
Passwords and keys the agents and apps need live in K2's encrypted vault on your server, not in chat or in the code.
Who can log in
Each app has its own logins and roles
A person can have a different role in each app, and each login sees only what its role allows.
Access ends when you say so
Turning off someone's access ends their session right away.
Customers and subcontractors can log in too
Each one sees only their own job. On current K2 servers running Linux, a login can be limited to one folder.
How a change reaches your team
- Step 1
Someone asks
A staff member asks the agent team for a change in plain words.
- Step 2
An agent makes it
The agent that knows that part of the app does the work.
- Step 3
Your rule decides what happens next
Straight in, wait for your approval, or try it on a test copy first.
- Step 4
It's your code
Any developer you choose can review a change, and roll it back if it's wrong.
Who you call when something breaks
Your agent team is the first stop
Most everyday problems and changes are a message away.
People, when you want them
Email hello@alakazamlabs.com or book a call. We typically respond within 24 hours during business days. Expert help is there whenever you want it.
Ongoing care, if you want it as a standing service
Our people check on your agent team regularly, handle the fiddly settings, and know your software and your team. It's a paid service, priced by the size of your agent team.
New features are their own project
We quote them before we start.
What we don't have yet
A full log of every agent action
Today the log covers what the apps do, not every action an agent takes.
SOC 2
We don't have SOC 2 certification today. If your company requires a SOC 2 certified vendor, we're not the right fit yet, and we'll say so on the first call.
App walls on servers that don't run Linux
The walled-off app runtime runs on Linux servers (K2 0.46.1 or later). Ask us about your setup on the call.
Questions IT teams ask
Is software written by AI agents safe to run?
It should meet the same bar as any software you run. On current K2 servers running Linux (0.46.1 or later), each app runs walled off, reaches only what you approved, and every app call is logged. Our people check the agents' work during the build, and you set the rules for changes after launch.
Who can see our data?
Your software and data run on the server you choose. On your own server, it stays with you. The agents send prompts to the AI provider on the plan you choose, under that provider's terms.
What happens if Alakazam Labs goes away?
You own the code and the data, and K2 is free to download. On your own server, your software and your agent team keep running, and any developer can work on the code.
Can we review the code?
Yes. It's your code. Any developer you choose can read and review it.
Do you have SOC 2?
Not today. If that's a hard requirement, we'll tell you on the first call.
Ask us the hard questions
Bring your IT checklist to a free call. If we can't meet it, we'll tell you.
Or email hello@alakazamlabs.com