The answers a review asks for, on one page.
Three kinds of answer live here. Where a claim can be checked without us, we say how to check it. Where it rests on a contract, we name the contract. Where we don't hold something, we use the word "no".
Last reviewed 26 September 2026
One firm, and its name is on the door.
- The firm
- FuseLoom — an AI consultancy. Six services: AI agents & automation, applications & solutions, AI strategy & roadmap, data & insight, digital transformation, AI enablement.
- Legal entity
- to confirm · registered name & number
- Where we are
- Kirresh AutoMall Complex, 148 Mecca Street, Amman, Jordan.
- Who answers
- Basem Qreieyeh, CEO & Founder — hello@fuseloom.com. Correspondence in English or Arabic.
- Who does the work
- The people on the discovery call are on the team that builds it. If a phase needs a specialist, we name them before they start — never somebody you would only ever meet in a support thread, and never anyone working beside our contract rather than under it.
We ask for access. We don't ask for copies.
Your data quietly accumulating on somebody else's laptop is the part of a consulting engagement nobody writes into the contract. Here is how we keep it from happening.
You issue the access, and you can end it in a click. We work in accounts you create in your own systems, named to a person, with the narrowest permission the task allows. When the phase ends, you revoke them — you don't have to ask us to hand anything back first.
We build against the smallest real sample we can. Where an example only has to show the shape of your data, redacted or synthetic records do the job, and we'll say so rather than take the live export because it was offered.
We do not take a standing copy of your database, your CRM export or your customer list. What we hold during a project is the working material: notes, diagrams, prompts, code, and whatever sample was agreed in writing at the start of the phase.
Where that working material lives: On a machine with full-disk encryption that locks itself when it is left. The accounts that could reach any of it — Google, Cloudflare, the code host — each carry two-factor, with a separate password from a password manager rather than variations of one.
At the end, everything is handed over — code, prompts, accounts, documentation — and our copies of your material are deleted on request, with a written confirmation of what was deleted and when.
What a model sees, and what it is allowed to keep.
"Your data never trains AI models" is a sentence any firm can type. A sentence like that is only ever as good as the account it runs on. So: the account, the term, and the retention — the things a reviewer can check against a provider's own contract.
Business terms, not the consumer app
Client work runs on business API accounts, never on consumer chat products, where the terms and the defaults are different.
Anthropic’s API, on their commercial terms. The assistant on this site and any model work we do for you runs through it. Nothing of yours touches a consumer Claude account, where the terms and the defaults are not the same.
Training on your content is excluded by contract
Not by policy on our side — by the provider's own terms for that account type, which you can read without our involvement.
Anthropic’s Commercial Terms of Service, section B: “Anthropic may not train models on Customer Content from Services.” Effective 17 June 2025 — read it without us.
How long a prompt survives
Requests and responses sit in the provider's systems for a stated window and are then deleted. The window is theirs to publish, so we point at it rather than describe it.
Thirty days. Anthropic delete API inputs and outputs within thirty days of receipt, with narrow exceptions they publish. We have not arranged zero retention — that is a separate agreement with them and we do not hold one. If your work needs it, say so before we start rather than after.
We never turn your content into weights
FuseLoom does not fine-tune models on client material. Systems we build retrieve your approved documents at the moment of the question — your knowledge stays a file you control and can delete, not something baked into a model.
Only the fields the task needs
A model that answers a delivery question does not need a national ID number, so it never receives one. Identifying fields the task doesn't use are stripped before the request leaves your systems, and which fields travel is agreed in writing before anything is connected.
It stops rather than invents
An agent answers from your approved material or not at all. A human keeps anything involving financial records, a contract or a relationship. The agent stops there and asks.
No SOC 2. No ISO 27001. Not today.
We will not display a badge we do not hold.
Those audits exist to answer one question well: can this vendor be trusted with a copy of our data inside their building? That is the right question for a company that hosts your customers on its own servers, and the work those audits demand is the right work for them.
A consultancy is a different shape of risk. We do not host your data — we work inside systems you own, under credentials you issue and can revoke. So the questions that actually protect you here are: who has access, granted by whom and for how long; what is in the contract; who is named when something goes wrong; and which outside services touch anything. This page answers each of them, and a signed DPA makes the answers binding.
Where an engagement changes that shape — if we host something for you rather than deliver it into your own accounts — the arrangement is written into the statement of work, along with who is responsible for what. If a certificate is a hard condition of your procurement, say so on the first call. Better we both know at the start than at signature.
Everyone else who touches this.
These are the outside services involved in running fuseloom.com and our own correspondence. Client engagements may add providers; those are named in the statement of work.
| Provider | What it does for us | What reaches it | Where it is processed |
|---|---|---|---|
| Cloudflare Their trust hub & data terms ↗ | Hosts the site and sits in front of it as the network that serves every page. | Your IP address, the page you requested, and the ordinary details every browser sends. | Cloudflare's global network — served from the location nearest the visitor. |
| Google Workspace Their data processing terms ↗ | Runs our email — everything sent to and from hello@fuseloom.com. | Your message, your address, and anything you attach to it. | No region policy is set, so Google hold it where they hold it — in practice the United States. If your work needs it kept in Europe, that is a setting we would change before starting, not after. |
| Resend Their data processing terms ↗ | Delivers the two emails the site sends: your request as it reaches us, and the acknowledgement that goes back to you. | What you typed into the form: your name, the email address or mobile number you gave, the time of day that suits, and your note. | North Virginia, in the United States — the region this domain is set up in, which anyone can check from its mail routing record. Ireland, Tokyo and São Paulo are the alternatives; moving to one means re-adding the domain and changing a DNS record, so it is something we do before a job starts rather than during one. |
| Anthropic Their commercial terms ↗ | Writes the replies the assistant in the corner of this site gives you. | What you type into that panel, and the conversation so far. Nothing else about you — not your name, not your address, not who you are. | On Anthropic’s API under commercial terms, not a personal Claude account. Inputs and outputs are held for up to thirty days; we have not arranged zero retention, and say so above rather than leave it implied. |
| Google Fonts What they receive, in their words ↗ | Serves the typefaces this page is set in. | Your IP address and basic browser details, at the moment the font files are fetched. | Google's global network. |
Paper first, if that's how you work.
Ask at hello@fuseloom.com and the current drafts come back by email — before you send us a document, a login, or anything at all. Send your own templates instead and we'll mark plainly what we cannot sign.
Mutual, so it protects your side and ours. Signed before the conversation gets specific enough to matter.
The master agreement: liability, ownership of what we build, confidentiality, how either of us ends it. Signed once, then every project hangs off it.
One per phase. What gets built, what you keep, what it involves and how that scope was arrived at. Each phase ends at a door you can walk out of.
The data processing agreement: what we may process, on whose instructions, with which sub-processors, and what happens to it at the end. It makes the table above enforceable instead of merely published.
The name you write to, and what happens next.
Security contact: Basem Qreieyeh — hello@fuseloom.com. Put "Security" in the subject line and it goes to the top of the pile. It reaches a person, not a ticket queue.
What you get back: an acknowledgment in writing from that person. Then, as we learn it — what happened, what we know for certain, what we don't know yet, and what we have already changed. We keep writing until it is closed, and we do not wait until we have a tidy story to send the first message.
If the cause sits with one of the services named above, we say which one and point you at their status page and their notice, so you are reading the source rather than our account of it.
Notification timing is a contract term, not a promise on a web page. It lives in the DPA, where it is enforceable — and it is one of the things we would rather agree before an incident than during one.
Found a flaw in this website itself? Same address.
The cross-border question, before your counsel asks it.
Almost every data protection regime — Jordan's Personal Data Protection Law, the GDPR, and most of what sits between them — requires an organisation to be able to say where personal data goes and why it is allowed to go there. Wherever you are, your counsel will ask us. Here is the answer in advance.
This website and its contact form use services that process outside Jordan, where our office is. They are named in the table above, with what reaches each one — so you can check the list against your own rules rather than ours.
In an engagement, the default is that your data stays where it already is — inside your systems, under your accounts. What crosses a border is the specific request a system has to send to do its job, not your records.
If your policy or your regulator requires processing to stay inside a particular country or region, say it on the first call. It changes the design: which providers are available, whether a hosted model can be called at all, and sometimes whether the honest answer is a simpler system with no model in it. That is a real constraint to design around, not an objection to talk you out of.
This describes how we work. It is not legal advice — what your organisation is obliged to do is your counsel's call, and we will give them whatever detail they need to make it.
Checkable from your own browser, right now.
Every row below is a response header this site actually sends, copied word for word from the file that sets it. Paste fuseloom.com into any public header checker and read them for yourself — we cannot edit what it reports.
| Header | What it is set to | What that means |
|---|---|---|
| Strict-Transport-Security | max-age=31536000; includeSubDomains | Your browser refuses to talk to this site unencrypted, and keeps refusing for a year after each visit — across every subdomain. |
| Content-Security-Policy | default-src 'self'; script-src 'self' 'unsafe-inline' https://static.cloudflareinsights.com https://challenges.cloudflare.com; style-src 'self' 'unsafe-inline' https://fonts.googleapis.com; font-src 'self' https://fonts.gstatic.com; img-src 'self' data:; connect-src 'self' https://cloudflareinsights.com https://challenges.cloudflare.com; frame-src https://challenges.cloudflare.com; object-src 'none'; base-uri 'self'; frame-ancestors 'none'; form-action 'self' | Names every origin allowed to serve code, styles, fonts and images, blocks plug-ins and framing, and stops any form here posting anywhere but back to us. Its one loosening is 'unsafe-inline' — this site's own styles and scripts are written into the page rather than fetched as files, so the policy has to permit inline code, which weakens the protection against script injection. We print the policy in full rather than reprint the flattering half of it. |
| X-Frame-Options | DENY | Nobody can load this page inside their own frame and dress it up as theirs. |
| X-Content-Type-Options | nosniff | The browser never guesses what a file is. It takes our word or it refuses it. |
| Permissions-Policy | camera=(), microphone=(), geolocation=(), interest-cohort=() | Camera, microphone and location are switched off for the whole site — nothing here is even able to ask — and the page opts out of being used to sort you into an advertising cohort. |
| Referrer-Policy | strict-origin-when-cross-origin | When you follow a link away, the next site learns you came from fuseloom.com — not which page you were reading. |
Something on this page not precise enough for your review? Write to hello@fuseloom.com and tell us which line. If you're right, we'll fix the line — and if the answer is "we don't have that yet", it will say so here in those words.