Tickets your team understands.
Resolutions your tools draft.

AtlasDesk is modern ITSM with natural-language ticket creation and an AI Resolution Copilot that drafts fixes from your past tickets and knowledge base. In production today at a liberal arts college.

TKT-2841atlasdesk.app / tickets / TKT-2841
TKT-2841 · reported by m.alvarez@wooster.edu · 2m ago
Projector in Kauke 243 won't mirror from MacBook — HDMI handshake fails after sleep
Open · Tier 1
COPILOTDrafted resolution3 sources · TKT-2617, KB-44, KB-51
[ HERO ANIMATION — replaced at build time by HeroCopilotAnimation ]
Resolution Copilot

Your tier-1 queue, suddenly tier-2.

When a ticket lands, Copilot runs semantic search across every past resolution and KB article, drafts a candidate fix with citations, and hands it to a technician for a single click of review.

Embeddings live in pgvector and are generated locally with transformers.js; drafting is done by Gemini with retrieved context only. No training on your ticket history, no data leaving your tenant.

TKT-2847 · Copilotresolution-copilot / draft#1 · 14s elapsed
45 min 14 sec
Median time from ticket creation to drafted resolution, Wooster IT, Q1 2026.
3 sources · cited
Every draft cites the past tickets and KB articles it drew from. Nothing lands without receipts.
Tech in the loop
Copilot drafts, a technician reviews and sends. The Copilot never closes a ticket on its own.
Natural Language Booking

“I need a projector for Tuesday's lecture.” That's it.

Type a request the way you'd ask a colleague. AtlasDesk parses the ask, finds an available asset, holds the reservation, and if a hardware visit is needed, drops the appointment directly onto the technician's calendar.

Rooms, carts, loaner laptops, AV kits — anything in the asset inventory is bookable the same way, with the same conflict rules.

Reservations · newreservations / parse + book
AtlasDesk · Submit a request
Describe your issue
Hardware Team · This week
MONTUEWEDTHUFRI
9 AM
10 AMLab inspection10:00 AM
11 AM
12 PM
1 PM
2 PM
3 PM
4 PM
Everything you expect

And the boring parts done right.

Kanban and list views. Asset inventory with QR tagging. A week-at-a-glance reservations calendar. The table stakes, without the Confluence-era feel.

TKT-2841
TKT-2839
TKT-2835
TKT-2830
TKT-2826
TKT-2820
TKT-2818
TKT-2815

Tickets

kanban · list · sla

Assets

inventory · qr · lifecycle
Week of Apr 205 bookings
09
10
11
12

Reservations

rooms · carts · av
Role-based from the ground up

Tier 1 to Tier 4. STA to Director.

Every view, action, and API route respects the same role graph. Student technicians see only what they should; directors see the whole organization. The model was designed for higher-ed IT, and it scales cleanly to mid-market.

Built for engineers who inherit this later

Boring infrastructure, on purpose.

Nothing exotic in the stack. Postgres where Postgres fits, SQL Server where the customer requires it, and a single codebase that ships to both single-tenant on-prem and multi-tenant SaaS. The interesting work lives above the infrastructure, not inside it.

NestJSapi · workers · auth
Next.jsapp router · rsc · edge
PostgreSQLprimary store · multi-tenant
SQL Server 2022on-prem / wooster
Prismaorm · migrations
pgvectorembeddings · ann
transformers.jslocal embeddings
Dockersingle-tenant deploys
SAML · OIDCsso for both tenancies
One codebase. Single-tenant on-prem (Wooster) and multi-tenant SaaS (atlasdesk.app) ship from the same repository.
Still under active development.