Give your agent real capabilities — safely
Your qbit agent is powerful out of the box. Toolsets let you extend it with capabilities you choose — a calculator, a data processor, a device controller — packaged as a signed, sandboxed toolset that runs on your own appliance.
Each toolset is a small, self-contained program that your agent calls as a tool. It runs in a strict sandbox: no access to your files, your network, or your other data unless you explicitly grant it. You stay in control of what your agent can do.
Sandboxed by design. Every toolset runs in an isolated sandbox — it can't touch your files, your network, or your other data unless you grant it. Even a toolset you install from a stranger can't reach beyond its sandbox. Your hardware. Your rules.
Why sandboxing matters
Most AI tools run with broad access to your machine — or worse, on someone else's cloud where you can't see what they touch. Toolsets are different. Each one runs in a strict, isolated sandbox with only the permissions you grant it.
That means you can install a toolset from any publisher and know it can't read your files, reach your network, or touch your other data. It does its job, returns a result, and stays contained. If a toolset misbehaves, the sandbox stops it — your data and your appliance stay safe.
And because the sandbox runtime is open source, "can't touch your files" isn't a claim — it's a file you can open. Audit the sandbox runtime on GitHub →
Leverage toolsets built by others
You don't have to build every capability yourself. The qbit marketplace is full of toolsets from other publishers — specialized calculators, data processors, device controllers, domain-specific tools. Browse, install, and let your agent use them.
Because every toolset is signed, reviewed, and sandboxed, you can install one from a publisher you've never heard of and trust it to stay contained. You get the benefit of someone else's expertise without the risk of running their code with broad access to your machine.
Someone else's expertise, on your terms. Install a toolset from the marketplace and your agent gains a capability you didn't have to build. Signed, reviewed, and sandboxed — so you can trust it even from a stranger.
How toolsets work
1. Find a toolset
Browse the marketplace for a toolset that fits your work — built by you or by another publisher. Each one is signed by its publisher and reviewed before it's listed, so you know what you're installing and who made it.
2. Install it on your agent
Pick the qbit agent you want to extend and install the toolset. It lands on your appliance, verified and ready. Installing a toolset is your consent for your agent to call it.
3. Your agent calls it as a tool
When a task needs that capability, your agent invokes the toolset. It runs in the sandbox, does its job, and returns a result — metered and recorded, so you can see exactly what ran.
What's included
- Sandboxed by default: Every toolset runs in an isolated sandbox with only the permissions you grant — no broad access, no surprises
- Signed and verified: Every toolset is signed by its publisher and checked before install — no tampering, no impersonation
- On your hardware: Toolsets run on your qbit appliance, not a third-party cloud
- Metered and transparent: Every invocation is recorded, so you can see what your agent ran and when
- Full lifecycle control: Install, run, deactivate, or archive toolsets from your dashboard
- Your consent: Installing a toolset is your choice — your agent only calls what you've installed
Publishing cost
Installing and running toolsets is free. Publishing a toolset to the marketplace is a paid action:
| Action |
Price |
Free / day |
| Toolset create |
$0.01 |
5 |
| Toolset publish |
$0.02 |
5 |
| Toolset attest |
$0.02 |
5 |
Free users can still install and run toolsets on their appliance. Publishing a toolset to the marketplace requires a paid call. See the full pricing page for all action prices and free allowances.
Use cases
- Extend your agent: Add a capability your agent didn't have — a specialized calculator, a data formatter, a domain-specific processor
- Leverage the marketplace: Install a toolset built by another publisher and gain a capability you didn't have to build — signed, reviewed, and sandboxed
- Control devices: Install a toolset that lets your agent read and actuate your IoT devices, within the limits you set
- Keep it private: Run toolsets on your own hardware, with your data staying on your appliance
- Trust what you install: Signed, reviewed toolsets mean you know what your agent can do — and what it can't
- Run third-party code safely: Install a toolset from any publisher and know it's contained in its sandbox — it can't reach your files, network, or data