XANDITE
Your pricing logic, vendor calls, and service-area judgment — running as a private agent, not a chatbot. Nothing leaves the tenant.
Every quote routes
through one person.
Staff at owner-operated trades businesses field pricing exceptions, vendor calls, and service-area edge cases all day — and most of that judgment lives in one head. When the owner isn't reachable, the job stalls or the answer gets guessed.
Grounded in what the business already knows.
No general-purpose model, no public knowledge. The agent only ever answers from documents this specific business uploaded — pricing sheets, vendor terms, service-area rules.
- 01
Upload
Pricing sheets, SOPs, and vendor terms go in — PDF, DOCX, TXT, or Markdown.
- 02
Chunk & embed
Documents are split and embedded locally, tagged to that tenant only.
- 03
Retrieve
A staff question triggers a tenant-scoped vector search — never another business's data.
- 04
Answer, cited
The agent responds with the retrieved passage cited, so staff can verify the source.
Enforced in code, not policy.
We don't ask you to trust a settings page. The isolation and data-handling rules that matter are structural — checked on every request, tested on every deploy.
Checked twice, every time
Your data is filtered by your business's ID at the database level, then checked again by the application before any answer goes out — a bug in one layer can't expose another business's information.
A safeguard nobody can quietly disable
The check that keeps one business's data from leaking into another's runs on every single deploy. It's built to stop the deploy if it's ever removed, not just flagged.
No real data to public AI, ever
If a business's documents are real (not test data), the system physically blocks them from reaching any public AI service — not a setting someone could leave off by mistake.
Every action is on the record
Every question asked and every change an owner makes is written to a permanent log they can review anytime. Nothing happens invisibly.
By the numbers, not by the pitch.
Engineering targets, not marketing claims — this is what the architecture is built against.
For businesses where judgment is the product.
Xandite is built for owner-operated service businesses with high call volume and judgment-heavy decisions — not general offices.
Load & disposal pricing
“Can the fridge ride with the couch and mattress, or does it need its own trip?”
Vendor & parts logic
“Which vendor do we use for this part when the usual supplier is backordered?”
Service-area exceptions
“Do we charge the after-hours rate for a call booked at 5:45pm?”
Warranty edge cases
“Is this unit still under the manufacturer warranty, or ours?”
Treatment sequencing
“Can we treat for both in one visit, or does that void the guarantee?”
Seasonal quoting
“Does the fall cleanup rate include the second pass if leaves keep dropping?”
Why we're building this.
Too much of how a small business actually runs — the pricing logic, the vendor relationships, the reasons a job gets handled one way instead of another — exists only in one person's head. That knowledge doesn't scale. It doesn't survive a slow week to onboard a new hire, a busy season, or a day the owner can't be reached.
Xandite is our attempt to fix that: turn what an owner already knows into something their whole team can draw on, without ever handing that knowledge to a public AI company. We're building this hands-on, one business at a time, not as a self-serve product — because the businesses we're building for deserve something shaped around how they actually work, not a generic template.
We're early. If you run a service business and you're curious whether this could work for you, we'd genuinely like to understand how your business runs before we tell you anything more about ours.
What staff actually see.
A private conversation with the tenant's own agent. The greeting, the ambient glow, and the grounded citations below are real, shipped portal behavior — reskinned here, not invented.
Answers are grounded in the documents uploaded to your workspace, with sources cited below each reply.