Ownership & Security
Built inside your walls.Gated at every write.
An agent that touches your customers and your money deserves more scrutiny than any other software you buy. Here that scrutiny starts with architecture: there is no Vulc cloud to trust, because there is no Vulc cloud. The system runs in your environment, behind gates you set. This page is the discipline every build ships with.
- Your environment
- Least privilege
- Approval gates
- Audit trail
- Loud failure
- Clean handover
The discipline
Architecture first.Promises second.
Your environment
The system runs on your accounts and your infrastructure. Your data stays where it already lives, under contracts you already signed. If Vulc disappeared tomorrow, everything would keep running.
Least privilege
Each agent gets scoped credentials per integration, sized to its job and nothing more. The support agent can draft a refund. The research agent cannot write anywhere at all.
Approval gates
Every write action stops for your sign-off. You decide which routine actions earn standing approval, where the money thresholds sit, and what stays off-limits entirely.
Audit trail
Every draft, decision and send is logged: which agent acted, on what, with what context, and who approved it. When you want to know why something happened, the answer is on record.
Loud failure
Failure states are designed before launch, not discovered after. When something breaks, the alert names the failing step and the system holds instead of guessing. Nothing fails silently.
Clean handover
The repository sits in your GitHub organisation from the first commit, so there is no transfer step. Our access is granted by you and revocable by you, and after handoff nothing standing remains.
Autonomy, per action
As autonomousas you decide.
Autonomy is never a switch for the whole system. Each action an agent can take sits at one of four levels, set with you in the design stage and moved as trust builds. The agent permissions map in the pack lists every row.
The same rule shapes integrations: an agent gets the narrowest credential the job needs, per system, granted by you and revocable by you. If the required access and endpoints are available, we can integrate the system into the build; we verify that before scope is locked.
The integration surfaceRead only
The agent reads and reports. It never writes anywhere. Research and monitoring roles usually live here.
Draft, you send
The agent prepares the reply, the change or the report. Nothing leaves until a person approves it.
Standing approval
Routine cases below a threshold you set run on their own, after proving themselves against real cases. Everything above the line still stops.
Never
Some actions stay human by design: pricing calls, legal threats, anything you name. The agent routes them and steps back.
Testing, expanded
Nothing meets productionbefore launch.
Stage five of every build is testing: permissions, gates, failure states and volume, tested before the system touches a real customer. Testing ends the only way it can: with your sign-off on the go-live.
Test report · illustrative
- permissions · scoped per integration · pass
- gates · every write action covered · pass
- failure drill · alert fired, named the step · pass
- volume test · run against real load · pass
Production go-live
tests passed, checklist documented
Needs your call
approval-first. nothing ships without you.
The model account
Your key. Your bill.Your agreement.
Most AI work runs model calls through the agency's account. Your prompts flow through their organisation, you cannot see the spend, and your data sits under their agreement with the provider. We do it the other way around. You create your own Anthropic account and API key. Agent calls run on your organisation, billed to you and visible to you, under your own agreement with Anthropic.
- Anthropic does not train models on Customer Content from paid services.
- You own the Outputs.
- For the EEA, UK and Switzerland the contracting entity is Anthropic Ireland, Limited.
- Standard API retention is 30 days.
The stack
Five platforms.All of them yours.
You create each account under your own billing and add us as an admin collaborator for the build. Vercel and Supabase run on AWS, so your data sits in a Postgres database on AWS infrastructure, in the EU region your project is created in. We use the hyperscalers through managed platforms rather than operating them directly, and your contracts are with the platforms, not with us.
Vercel
Your account
Application hosting. Holds: application runtime and logs.
Runs on AWS
Supabase
Your account
Postgres database and auth. Holds: system data and the audit log.
Runs on AWS, in the EU region your project is created in
Inngest
Your account
Background jobs. Holds: job payloads in transit.
Anthropic
Your account
Model inference. Holds: prompt context per request.
Anthropic Ireland, Limited for the EEA, UK and Switzerland
GitHub
Your account
Code repository. Holds: code only, no customer data.
Your organisation, from the first commit
Vulc
Our side
Your contact details and your build documentation. During the build, the admin access you granted. After handover, nothing standing.
Standard builds add nothing else. If yours needs error tracking, email delivery or monitoring, it is named in your architecture document rather than assumed here.
Certifications, honestly
We wear no badges.You inherit yours.
Vulc holds no security certifications and rents you no infrastructure, so there is nothing of ours to audit. Your deployment inherits the posture of the platforms it runs on, under your own contracts with them. Every build is documented at handoff: a permissions map per agent, the data flows, and the runbooks for incidents and revocation. If a vendor shows you a badge wall for a system that runs in your own account, ask what the badges are for.
Vulc is the trading name of ECOMFORWARD LLC, a Delaware company, operated by Maarten Klijnsma from Amsterdam. A Dutch entity is planned and contracts will move to it once it exists. Where a build touches personal data, a data processing agreement is signed before we access anything.
Security questions: security@vulc.ai
The pack
Ask for the documents.You get them.
Six documents back this page. The overview is ready to read today. The rest are filled in for your build and signed before anything touches production. Ask and they arrive, no call required first.
Security overview
This page in document form, with the subprocessor list and the common questions answered.
Architecture overview
Per build: deployment model, environments, access, secrets, logging, change management.
Agent permissions map
Per build: what each agent may read, draft and write, with the gate and threshold on every row.
Data flow map
Per build: what data moves where, what leaves your systems, and what is deliberately not sent.
Incident response runbook
Per build: how to stop an agent, contain, assess, roll back and report. Yours at handover.
Data processing agreement
Signed before we touch a system of yours that holds personal data.
or write to security@vulc.ai