One tenant, always
Your own copy. There is no shared database holding your records next to someone else's.
KickSide/Use cases/Regulated industries
Use AI without your data leaving the building
Legal, finance, health, insurance, EU. You want what AI can do and you can't hand your client list to a vendor. KickSide runs on infrastructure you control, one company per copy, with a record of everything an agent did.
Where everything sits
One boundary, and what is allowed to cross it
This is the whole architecture on one page. Print it and hand it to whoever owns your vendor review.
Your infrastructure your server, your cloud account, or fully offline
KickSideOne company per copy. No shared environment.
InsideAgentsStart with no access. Checked on every attempt.
InsideYour recordsContacts, deals, documents, history.
InsideAudit recordWhat ran, for whom, against which record.
InsideNothing in here is shared with another customer
What crosses the boundary
A model callYour prompt out, the answer back, on your own key and your own account. We never see the key.
You chose itSoftware updatesOnly when you ask for them. Nothing installs itself and nothing phones home.
You chose itYour recordsNever. Not to us, not to another customer, not as training data, not aggregated, not anonymised.
NeverExactly two things leave, and both because you chose them. Everything a reviewer usually worries about stays on the left.
Never training data
A written term in the contract, not a line in a policy page that can be edited later.
Your model keys
You connect the provider and pay them directly. We don't hold, proxy or mark up your usage.
Exportable record
Read it in your own environment, any time, without asking us to run an export.
Your model, your key
The model can run on your own hardware
Here a self-hosted GPU provider is connected on an internal address, with a default model picked for each capability. Ollama, LM Studio and vLLM sit next to hosted providers, and the step can be finished later under System, Models.
The record
What your compliance team will actually read
Every run writes one of these, in your environment, where your own people can query it. This is a single entry from an overnight automation.
Audit record read-only · exportable
- When
- 06:00:12 · 14 March
- What
- Deal-health monitor, scheduled run
- As
- Automation, running as the person who set it up. No borrowed credentials.
- Read
- 12 contacts, 5 deals. Each record listed by id.
- Model
- Your provider, your key. Prompt and response length logged.
- Did
- Wrote one inbox item. Sent nothing to anyone.
- Where
- ops-01.yourcompany.internal
Module permissions set by you
Read contacts and dealsThe records it needs, and nothing beside them.
AllowedWrite to the approval inboxDrafts only, never outbound.
AllowedSend emailNeeds a named person, every single time.
Needs a personRead files on diskOutside the platform.
DeniedReach the internetNot granted to this module.
DeniedIllustrative entries. The shape and the fields are real; the values are not from a customer.
External agents
An outside AI client connects with a token you issue
Connecting Claude or another MCP client needs a token issued from your copy. The setup token carries your full account access; issue narrower ones under MCP, confined to named records, knowledge bases and workflows, with an expiry. Each token is shown once and can be revoked.
What your security review will ask
The answers, in plain terms
Most tools fail this review on the first question. This one is built to be read by the people asking it.
Where does the data go?
Nowhere. It stays on your servers or in your cloud account. Nothing passes through ours unless you choose managed hosting, and then it stays in your own private environment.
What can an agent touch?
Only what you allow. A new agent starts with no access to files, the internet, or other parts of the system, and every attempt is checked against your rules at the moment it is made, not once at setup.
Is there a record?
Every action is logged: what ran, for whom, against which record. Your compliance team can read it directly, in your own environment, without asking us for an export.
Is anything shared with other customers?
No. One company per copy. There is no shared environment and no multi-tenant database holding your records alongside somebody else's.
Do you train on our data?
No. Your data, private modules and synced history are never used to train models. Not aggregated, not anonymised. That is a written term, not a policy statement.
Which AI models, and who holds the keys?
You do. You connect your own model provider and pay them directly. We never see your keys and we don't resell or mark up model usage.
Can we get our data out?
Any time, in full. If you ever leave, your copy stays readable for thirty days for export. History sync is off by default, and you can switch it off and delete what it stored whenever you like.
Is there a DPA?
Yes, for any deployment where we host your data. It comes with the first reply rather than after the fourth call.
What we'll send
The paperwork your reviewers will want
Ask once and this arrives together, before anyone books a second call.
Enterprise Addendum
Uptime commitment, service credits, response times, and the liability terms your legal team will want to mark up.
Data Processing Addendum
For any deployment where we hold your data. Sub-processors, retention, deletion, and transfer terms.
KickSide Cloud Terms
The hosted service. Replaced by a signed Platform Licence for companies that need one.
KickSide Community License
The licence the software itself ships under. Summarised in plain language here.
Architecture overview
How permissions, logging and isolation actually work, written for an engineer rather than a buyer. On request.
Being straight with you
What we don't have yet
SOC 2 and HIPAA certification
We have the architecture those audits look for, and we don't have the certificates. We will tell you that on the first call rather than the fourth. If certification is a hard gate for you today, we are not ready for you yet, and we would rather say so than waste a quarter of your time finding out.
Signed, verified packages
Modules from our Hub are not yet cryptographically verified, so a module from any source installs and shows as unverified. If your policy requires signed supply-chain artefacts, that is on the roadmap and not shipped.
Where to start
Enterprise, sold by a person.
This isn't a self-serve tier. Tell us what you're trying to do and where it has to run, and we'll come back with a plan and the paperwork.
- Your own private cloud or a fully offline deployment
- Single sign-on, roles and audit
- A named engineer
- Uptime commitments, service credits and response times, set out in writing