SERVICEORBIT
From a few words
to a routed ticket.
ServiceOrbit is a multi-tenant AI service desk. A requester chats, talks, or emails in plain language — the engine detects intent, grounds in policy, parses the evidence, checks eligibility, and drafts a complete, routed ticket. No forms.
It covers the whole lifecycle — intake, approval, fulfilment, archive — with an AI assessment for every approver and SLA clocks that keep their own time. And a Service Engine that authors new catalogue services from a department's real documents.
Interactive demo · five sample tenants · no sign-up to look around
ServiceOrbit · Shared Services Assistant
Live intake · builds the ticket as you talk
Ticket #SO-4821 · drafted
AI recommendation: Approve — all 3 policy criteria met, evidence authentic.
SLA attainment
96% ↑ 3%
No forms
The assistant builds the request itself — the requester just describes the need.
Every field traced
Each value on the ticket links back to the policy clause or document it came from.
AI + human
The engine knows when to decide and when to route to a person for judgement.
INTAKE, IN PLAIN LANGUAGE
Your team stops filling in forms.
A requester describes what they need — by chat, by voice, or by email — and the assistant does the rest: it finds the right service, asks for exactly what's missing, reads the attachments, and hands a clean, decided ticket to the approver. The work moves; nobody wrestles a portal.
THREE WAYS IN
Chat. Voice. Email. One engine behind all three.
Whichever way a request arrives, the same intelligence runs underneath — intent detection, policy grounding, document parsing, and a drafted ticket at the end.
Chat
RequesterA proactive assistant identifies the right catalogue service by purpose and audience, asks one question at a time, and builds the request itself. The requester never fills in a form.
Voice
RequesterA live speech-to-speech agent (OpenAI Realtime over WebRTC) that builds the ticket as you talk — natural conversation, structured outcome, ephemeral keys minted server-side.
Inbound department email is auto-triaged into a routed, AI-drafted reply — the same intent detection and policy grounding, applied to the channel your requesters already use.
THE FULL LIFECYCLE
Eight stages, intake to archive.
The engine injects an LLM at every point where it adds value — and stops where human judgement belongs. This is the orchestration map the platform actually runs.
01
Converse & detect intent
Plain-language chat, voice, or email. The assistant identifies the right service by purpose and audience.
02
Ground in policy
Pulls the service’s policy clauses, rules, exclusions, and the requirements checklist.
03
Parse documents
Extracts structured fields from PDF/image uploads via vision — in-memory; only extracted fields persist.
04
Evaluate eligibility
Checks each criterion against the requester profile and the submitted evidence.
05
Decide
Proceeds, routes to a human approver, holds for a fix, or declines — and knows when not to auto-decide.
06
Draft the ticket
A legible, structured ticket where every field traces back to its source.
07
Route the workflow
A persisted AI assessment per approver — recommendation plus per-criterion verdicts judged against the actual policy.
08
Track SLA & status
Business-hours SLA clocks that pause on “waiting for requester,” proactive updates, and outbound Slack notifications.
SEE IT RUNNING
The actual product, up close.
Screenshots from the live ServiceOrbit demo — the orchestration map, the approver console and its queue, and the requester's status page. The demo runs the Knowledge Unit tenant; the engine is the same for every desk.
MEASURE, THEN IMPROVE
Run the desk by the numbers — and let the engine improve it.
Intake and decisions are only half of it. ServiceOrbit also measures the operation and proposes its own improvements — costed, and governed. Both views below are from the live demo.
THE SERVICE ENGINE · THE DIFFERENTIATOR
It writes its own catalogue.
Most service desks make you author every service by hand. ServiceOrbit reads a department's real documents and drafts a governed Standard Service Definition — policy, eligibility, evidence checklist, routing. Approve and publish, and the live assistant starts serving it the same minute.
- ● Onboard from documents — upload a real PDF/DOCX; the engine authors the service.
- ● Govern it — review the policy, eligibility criteria, and routing before anything goes live.
- ● Publish to D1 — the chat, voice, and email channels start serving it instantly.
- ● Improve it — costed CSI proposals per service, grounded in real demand.
Onboard → Author → Publish
Drop a department's real PDFs; the engine drafts a governed service against an 11-section template, then you approve & publish. The flagship demo tenant ships 9 hand-scripted services + 52 AI-authored services live in the serving catalogue.
THE GOVERNED STANDARD
Every service follows one lifecycle, one definition.
Behind the catalogue is a governed standard: each service moves through propose → author → govern → publish → operate → improve → retire, and conforms to one 11-section Standard Service Definition. From the live demo.
THE WORKSPACE
What approvers and requesters actually open.
The channels are how requests come in. These are the surfaces where the work gets decided, tracked, and improved.
Console
AI-drafted tickets, traced evidence, a persisted AI recommendation, and one-click decisions.
Dashboard
Live ops — backlog, SLA attainment, throughput, and a business-impact (ROI) band.
Track
An SLA ring, stage progress, and proactive AI updates on where the request stands.
Lifecycle
An orchestration map of all eight stages — where AI runs vs. where humans keep judgement.
Catalogue
The control plane that authors, governs, improves, and retires the services the channels serve.
Improve
Generates real, costed continual-improvement (CSI) proposals per service — grounded in live demand.
DECISIONS, NOT DATA ENTRY
Approvers decide. The engine did the legwork.
By the time a request reaches a person, it's already a clean ticket: evidence parsed, eligibility checked criterion by criterion, and a recommendation judged against the real policy. The approver reads, sanity-checks, and clicks — minutes of judgement instead of hours of triage.
MULTI-TENANT BY DESIGN
One platform, many desks.
The same engine re-skins its brand, accounts, catalogue, and AI framing per client. Tickets, metrics, and demo resets are tenant-scoped — one client's queue never touches another's.
Knowledge Unit
Higher education
Flagship — 9 scripted services + 52 AI-authored services live in the serving catalogue.
Vantage Group
Corporate shared services
Internal request management across a multi-department enterprise.
Northwind Bank
Banking
Policy-heavy intake with strict eligibility and evidence checks.
Meridian Health
Healthcare
Sensitive workflows with human judgement kept firmly in the loop.
Civic Services Authority
Government
Citizen-facing service requests with full audit and traceability.
Switch tenants from the login screen or the header — each one a fully separate desk on shared infrastructure.
See them in the demoTRY IT TWO WAYS
Demo mode is real. Live mode is really live.
The public demo runs fully-scripted scenarios with animated cascades on every channel — no key needed, nothing to break. Flip to Live mode and the same flows run on real model calls: conversational intake, document parsing, approver assessments, and service authoring.
Demo mode · default
Scripted scenarios, animated channel cascades, presenter script. Explore freely with no sign-up.
Live mode · real AI
Real model calls for intake, vision parsing, approver assessment, and authoring. Gated behind an unlock — it fails safe when locked.
HOW IT'S BUILT
Edge-native, and honest about the model.
Cloudflare Workers + D1
Next.js on the edge via OpenNext; serverless SQLite for tickets, events, and the published catalogue.
In-memory document parsing
Uploads are parsed in memory; only the extracted fields persist, and the raw files are discarded.
Zero-retention model calls
The LLMs ServiceOrbit calls operate under zero-retention API terms. We never train on your content.
Geofence-ready egress
An optional proxy routes model traffic for regions that geoblock OpenAI — Workers run near your users.
Same data commitments as the rest of Dendrites AI — see our data practices.
SEE IT FOR YOURSELF
Walk a request from words to done.
Open the live demo, pick a tenant, and watch a plain-language request become a validated, routed ticket across chat, voice, and email.