Your people don't want a ticket. They want it working again.
Relay answers in the tools your employees already use — chat, the desk line, email, SMS, Teams — resolves what it's allowed to resolve, and hands everything else to a person who already understands the problem. No forms. No reference numbers. No repeating yourself to a third person.
Ticket software is designed for the people counting tickets.
Every incumbent asks the person in trouble to do the desk's filing. Pick a category. Choose a priority. Quote a reference. Then wait, while a queue name that means nothing to them sits where a human's name should be.
She asks, in her own words.
Dana can’t reach the VPN and her board deck is on the share. She types one sentence on her phone. No form, no category tree, no reference to memorise — the request opens in the thread and the clock starts.
Type a sentence and send it. Relay answers from a published article, opens the request, and hands it to a person when it needs one — the same build the design system ships.
Starting the instrument…Try: type “the VPN says my certificate is not valid” · ask for something and watch the approval it names · open a request you already have.
The honest answer to “what does the AI actually do?”
Not everything should be automated, and pretending otherwise is how an AI service desk loses trust in week two. Relay climbs only as far as your guardrails allow, then hands over cleanly. Drag to see where a request stops.
A person arrives already briefed.
The server fix didn’t clear the client, so Relay escalates — with a handoff briefing stating what happened, what it tried and what it needs. Kofi’s work list shows the breach before he has read a word. Click the rows; press jk to move, a to assign, or ⌘K anywhere.
Not a screenshot and not a marketing render — the actual instrument, with its clocks ticking, its keyboard live and its command palette on ⌘K.
Starting the instrument…Try: open any of the ten rows — each is a real record with its own briefing, thread and suggested reply · jk move, ↵ opens, a assigns, n jumps to the worst clock · ⌘K runs anything · Ring the desk line answers a call that transcribes itself, and a caller we can’t verify shows what happens when the voice check disagrees.
The briefing is the first thing on the screen
Press The work item above. What happened, what Relay tried, what failed, what she needs and by when — written before you arrived, with its sources named. Zero orientation time, because reading a transcript is not the job.
Answer the call and it transcribes itself
Ring the desk line takes over the screen with who is calling, what they already asked today and how sure we are it is them — her voice matched her enrolled voiceprint at 0.94. Answer, and the call lands at the bottom of the same record, transcribing as it goes.
The suggested reply never steals focus
It is shown in full, cites the article it came from, and never sends itself. A suggestion you have to expand to read costs more than it saves.
The person, not the ticket.
Everything Dana has ever asked, in one stream — calls with transcripts, a screen share, the request still waiting on her manager. Health names its reason. Effort is a count, not a score. The third VPN failure in six weeks is flagged on arrival, because the fix for “again” is different from the fix for “once”.
Every channel in one stream — calls with transcripts, the screen share, the proactive message carrying its reason, the request still waiting on her manager — plus her device, her access and how sure we are it is her.
Starting the instrument…Try: open a call in the timeline and read the transcript · her devices, access and assurance are tabs on the same record · start a screen share from the header.
Never again, made literal.
A supervisor watched the whole thing live, sorted by risk, able to whisper without the requester seeing. And because the resolution wrote itself into a draft article, the next cached-certificate failure gets answered at 08:12 instead of 08:44 — or never happens at all.
Every live conversation sorted by clock rather than by owner, with the coaching drill-in: read the turns, whisper to the agent without the requester seeing, join, or move it to somebody free.
Starting the instrument…Try: open the worst clock · whisper to the agent on it · move it to somebody available.
06:45, three weeks later. The certificate monitor rotated the gateway certificate overnight. 214 people kept their access and nobody ever knew. Relay records it as a prevented event — not a ticket — so the numbers stay honest.
The desk gets quieter because it writes itself down.
AI is only as good as what it’s allowed to read. Relay answers from published knowledge only — so the knowledge lifecycle is a first-class surface, not a wiki someone abandoned in 2023.
Resolutions become drafts
Close a request with a resolution note and Relay drafts the article, carrying the source. Nothing publishes itself.
A person approves it
A supervisor checks the steps on a real device. Until then it stays out of every answer Relay gives.
Then it answers, at scale
Live in agent search and the portal. Staleness is tracked: “128 uses, not verified in 92 days.”
| Searched phrase | Searches | Best match | Last | |
|---|---|---|---|---|
| “zoom audio cuts out” | 14 | 31% | 2 hours ago | |
| “printer driver unavailable” | 9 | — | yesterday | |
| “teams status stuck on away” | 7 | 22% | 3 hours ago |
This screen is running: open Knowledge in the demo — search it, read an article, draft one from a gap and send it to review.
Nobody should have to learn where to ask.
Chat and portal
The default path. Relay answers first and hands over cleanly.
Voice
Relay answers the desk line, works the problem, and transfers with a briefing and a live transcript.
Email and SMS
Replies within the minute. Same queue, same conversation, no re-explaining.
Microsoft Teams
Answers, approvals and announcements where your company already talks.
Desktop tray
The desk number, your open requests, and known issues on your machine — before you report them.
Walk-ups
An agent logs it in two clicks, with duplicate detection built in.
One thing the tray does that nothing else does
Northwind IT1 844 737 1234
Desk open
An AI that can act needs limits you can read.
Published knowledge only
Never a draft, never an unverified article, never the open web. Every answer names the article it came from.
Actions are opt-in, per tenant
Your admins choose what Relay may do alone — reset, unlock, reissue, install. Anything that spends money needs a person.
Handover is configurable
After one failed attempt, two, or only on request. P1 always goes to a human, as does anyone who sounds distressed.
Everything is attributed
Machine-written text carries a distinct treatment throughout. You always know whether a person or a model said it.
Accessibility is ergonomics, not a compliance page
Your agents use this for eight hours a day. The same decisions that make it fast make it accessible.
ServiceNow, Jira Service Management and Freshservice route tickets between people.
They’re competent at that, and they were designed before an AI agent could resolve a request end to end. The difference isn’t a feature list — it’s which assumption the product is built on.
| Ticket-routing tools | Relay | |
|---|---|---|
| First responder | A person, once they get to it | Relay, immediately, on any channel |
| Escalation | A transcript and a category | A written briefing: what was tried, what failed, what’s needed |
| What the employee learns | Categories, priorities, reference numbers | Nothing. One sentence, wherever they already are |
| Status they read | “Tier 2 — pending” | “Marta from IT has this” |
| Agent input | Mouse-driven forms | Keyboard-first, command palette, every action keyed |
| Use of colour | Everything coloured, so nothing stands out | Rationed. Only time pressure is saturated |
| Knowledge | A wiki someone abandoned | Resolutions draft it; gaps are a queue; staleness is tracked |
| Prevented problems | Invisible — no ticket, no credit | Recorded and reported as their own category |
Most service desks measure tickets. Relay measures people.
Every one of these is a real component in the system, not a roadmap line. They exist because the thing that makes someone hate IT is rarely the fault — it’s the fifth time they explained it.
We count what it cost them
Touches, channel switches, follow-ups they had to chase, reopens. Four touches for one printer is visible before anyone complains. A score you can’t decompose is a score nobody acts on, so it’s always a count with the events behind it.
Three reopens is a person, not a metric
Healthy, watch, at risk — and never without the driver in words. A state with no reason gets ignored, and worse, gets argued with. The driver is the component; it is never a tooltip.
Three reopens in eleven days, all VPN. Worth a call, not a reply.
“It’s happening again” is one tap
No confirmation dialog, no required reason, no severity bump, no red. It returns to the same person with the history intact — because the fear that stops people reopening is having to explain it all over again.
“Pending” is deleted
Every waiting item names the party and the ask. Waiting on the employee is loud and imperative with the action beside it. Waiting on us is quiet, because that’s just work in progress. Waiting on a supplier carries the date we last chased.
Approve the profile on your laptop, then tell us it’s done.
You always know the words they’re reading
Internal state and employee-facing status sit together in every agent view. “Escalated” internally reads as “With a specialist” to them — mapped per transition, translatable, and yours to change.
The third time is flagged on arrival
“3× in six weeks” appears in the queue, on the conversation, and in the person view, linked to the earlier instances. Prominent enough to change what the agent does, without reading as an error — they didn’t do anything wrong.
Priced per agent. Your employees never count.
Relay charges for the people who work in it, not the people who ask it for help — so making self-service better never costs you more. End users are unlimited on every tier. Per agent, per month, billed annually.
The service desk itself, done properly.
- Incident, request, portal, knowledge base
- SLA engine and business calendars
- Email channel and AI chat deflection
- Announcements and broadcast
- Audit log, RBAC, custom fields
Every channel, and the endpoint agent. Voice, SMS, change management and SSO are all part of the tier.
- Everything in Lite
- Voice and SMS channels
- SSO and SCIM
- Desktop tray and unlimited remote support
- Change, problem, automation, custom reports
- Teams and Slack, custom domain and branding
Device context and on-call. The agent on each machine reports what it is, so your inventory stays live on its own.
- Everything in Core
- Entra and Intune, live device inventory and health
- On-call scheduling with AI triage-to-page and SMS accept
- Asset and licence management, monitoring and alerting
- Screen observation — Relay reads the screen, with consent
- AI QA and coaching, advanced approvals, Jira handoff
Relay acts on the endpoint — always with the employee’s consent, always inside your guardrails.
- Everything in Pro
- Action mode — Relay works the screen, not just reads it
- Endpoint remediation and Winget package management
- Proactive operations — problems fixed before they are reported
Every capability unlocked, nothing bundled — priced around how your desk actually runs. For a large desk, or one with light or unpredictable AI use.
- Everything in Autonomous, unlocked
- All AI, voice, messaging and screen use metered
- Prepaid credits, or hard stop instead of overage
- Volume commitments and committed-use terms
Allowances pool across your agents
The numbers above are per agent per month, but they are added together and spent from one pot. One heavy month on the desk line does not bill a specific person — and nothing carries over into the next month.
Employees are free, always
Five thousand people can raise, track and reopen without touching the bill. A desk that gets better at self-service should never pay more for it.
No charge per resolution
Charging for each thing the AI resolves punishes you for the outcome you bought. Allowances are denominated in messages, minutes and interactions — units you can count and reconcile.
Read-only seats cost nothing
Supervisors, auditors and managers who only need to look are free on every tier.
A hard stop is one setting
Allowance, then prepaid credits, then either overage or a hard stop — your choice. Choose hard stop and no invoice can ever surprise you: AI features degrade to the human queue and you are told.
Straight answers.
What makes Relay different from ServiceNow or Jira Service Management?
Those tools are built around the ticket. Relay is built around the person asking. An employee never sees a form, a category tree or a reference number they have to quote. An agent never opens a conversation cold — Relay writes a handoff briefing first, stating what happened, what was tried, what failed and what the person needs.
Does the AI answer from anything other than our own knowledge?
No. Relay answers from your published knowledge only — never a draft, never an unverified article, never a general model guess. Every answer cites the article it came from, so an agent can check it in one click and an employee can read the source.
What can the AI actually do on its own?
Only what your admins allow. Password resets, account unlocks, certificate reissues, licence assignment and software installs are typical. Anything that spends money or grants sensitive access routes to a person by default, and every automated action is logged as itself rather than as a ticket someone closed.
What happens when it can’t resolve something?
It hands over, and the handoff is the most carefully designed moment in the product. Before a human sees the conversation, Relay writes the briefing. Your agent starts with context instead of a transcript, which is where the handle-time saving actually comes from.
Does this work for employees who aren’t technical?
The portal is mobile-first and deliberately calm: one column, 44-pixel targets, sentence-case labels, and time stated in words rather than counted down. Employees never see an SLA clock, a queue name or a required reference number. Ownership is always shown as a person’s name.
Is Relay accessible?
Relay targets WCAG 2.2 AA throughout. Colour is never the only carrier of meaning, focus states are a designed surface rather than an afterthought, every screen is fully keyboard operable, live regions announce threshold crossings rather than every clock tick, and reduced-motion preferences are respected. This is ergonomics for people using it all day, not compliance theatre.
Which channels does it support?
Chat, voice, email, SMS, Microsoft Teams, the employee portal, a desktop tray agent, and walk-ups logged by an agent. Every channel lands in the same queue and the same conversation view — the channel changes the composer, never the workflow.
How does it handle multiple business units?
Relay is multi-tenant by design. SLA policies, routing rules, AI guardrails, channels, roles and portal branding are all per-tenant, and nothing is ever “global”. Employees see their own organisation’s name and mark, never Relay’s.
When can we get it?
Relay is in private development now, with design partners going live first. Join the list below and we’ll show you the product on your own ticket history, tell you honestly whether your volume suits it yet, and give you a date.
Bring us a week of your real requests.
Relay is in private development. We’ll show you what it would have resolved without a person, what it would have escalated, and exactly what the briefing would have said. Forty-five minutes, your data, no slides.