IT support & MSPs
Your AI IT support desk is still answering at 9:12 PM.
It opens the ticket with the client, the priority and the six users who can't work already in it. It pulls a nine-day-old ticket off hold with one text at 8:16. And when a caller describes a ransom note, it reads your client's containment steps and pages your on-call and your security lead in the same minute. Your desk is closed. The SLA clock is not.
Live in 24 hours · Works in your PSA, not beside it · Cancel anytime
R. Vance · on call
Northpoint Managed IT · after-hours board
Tonight · Tue Mar 30
4 open
- 7:00 PM
Ridgeway Accounting · Outlook offline
P3 - 7:30 PM
Halcyon Legal · backup job failed
NOC - 8:00 PM
Open
- 8:30 PM
Verdant Home Care · label printer down
P3 - 9:00 PM
Open
- 9:30 PM
Open
- 10:00 PM
Open
Waiting on client
- Fairmont Physical Therapy9 days · needs an after-hours window to reboot the file server
- Copperline Manufacturing4 days · needs sign-off on the replacement laptop quote
- Alder Grove Schools6 days · waiting on the ISP account number for the circuit move
Illustrative queue · synthetic tickets
Works with what you already run
It works in your PSA, not a copy of it
Every call and message becomes a ticket in the system your techs already have open — the client matched from the number, the priority set from that client's matrix, the transcript and the recording attached to the ticket rather than parked somewhere your team has to go looking. Where your PSA has an open API, we write the ticket, the status, the schedule entry and the note directly. Where it doesn't, we send a webhook or an email your PSA already parses, and your inbound rules stay exactly as they are. If a client's system is closed to us, we say so before you sign, and that ticket stays a structured message instead of a claim we can't keep.
- Tickets
- Opened with the client, contact, affected users and the caller's own words — not a three-word summary someone rewrites at 8 AM
- Dispatch
- Scheduled against your on-call rotation and that client's covered hours, so nothing lands on an engineer who isn't on
- Status
- Status moves with the ticket. One waiting on the client is marked waiting and stops escalating; it comes back out the moment the client answers
- Escalation
- Pages fired through your on-call tool with the client, the priority and the timestamp, and the acknowledgment written back to the ticket
Product names and logos are the property of their respective owners. Integration availability varies by system and plan.
One evening, five contacts
Watch it work the queue overnight
A phone call becomes a timestamped transcript. A web chat becomes a thread. An overnight job becomes a run log against the system it wrote to. Same agent, three different records — all of them filed before anyone unlocks the door.
Front desk
Warehouse system down · after hours
Inbound call · Bayline Logistics
R. Vance · tonight
- 7:30 PMHalcyon Legal · backup job failed
- 8:00 PMBayline Logistics · WMS · P2
- 8:30 PMVerdant Home Care · label printer
- 9:00 PMOpen
- 9:30 PMOpen
Definition
Plain answer
What is an AI help desk agent?
An AI agent for IT support is software that answers your help desk calls and live chats around the clock, handles common requests like password resets and status updates, and triages or resolves Tier-1 tickets on its own. It then does the work behind each request, opening, updating, and routing tickets and syncing everything to your PSA or ticketing software so your technicians only handle what truly needs a human.
6:12 PM to 7:55 AM
Six things that happen after the service desk goes home
One Tuesday at a forty-client MSP. Same evening as the board above, same tickets, same on-call engineer.
- 6:12 PM
A nurse at Verdant Home Care can't print visit labels.
Today
Voicemail, or a page to the on-call for something that is not a P1 and should never have woken anyone.
With CallSphere
Caller checked against the client's authorized contacts, printer model and error captured, P3 opened and scheduled for 8:30.
- 7:42 PM
Bayline Logistics loses the warehouse system at shift change.
Today
A page that says warehouse down, call Marisol, and an engineer who spends ten minutes working out what down means.
With CallSphere
Six users blocked, network confirmed up, P2 under their agreement, #284617 open and the on-call paged before the call ended.
- 8:11 PM
The Verdant printer starts working after someone power-cycles it.
Today
The engineer works the ticket anyway, or calls at nine to find out nobody needs him.
With CallSphere
The user says so in chat, the ticket is closed with the reason on it, and the 8:30 block comes back.
- 8:16 PM
A ticket that has waited nine days on the client gets its answer.
Today
It sits in waiting-on-client until someone runs a stale-ticket report on Friday.
With CallSphere
One text inside the client's contact hours, a yes at 8:19, off hold and into the block that just freed up.
- 9:12 PM
Kestrel Dental's practice manager finds a ransom note on the shared drive.
Today
An answering service takes a message. The on-call finds out when he next looks at his phone.
With CallSphere
The client's own containment steps read out loud, on-call and security lead paged together, acknowledged in 38 seconds.
- 7:55 AM
The service desk logs in.
Today
Seven voicemails, four tickets that need rewriting, and a P1 nobody can reconstruct.
With CallSphere
Four tickets with transcripts attached, three already scheduled, and a P1 with a timeline that starts at the first word of the call.
Scope
A dispatcher, not an engineer
The agent runs the front of your service desk. It does not touch a client's systems, and it is not built to.
Handles on its own
- Answers every call, chat and text on your after-hours line, and matches the caller to the right client tenant before anything else happens
- Opens the ticket in your PSA with the client, contact, users affected, what still works, and the caller's own description
- Sets priority from that client's matrix — it can raise a priority on new information, and never quietly lowers one
- Schedules against your on-call rotation and the hours that client actually pays for
- Pages the right person through your on-call tool and writes the acknowledgment time back to the ticket
- Chases the tickets stuck on the client, inside the contact hours each client sets, and takes them off hold when they answer
- Answers the calls that never needed an engineer: ticket status, scheduled time, what's covered, how to reach the portal
- Runs the whole call in the caller's language, in 57+ of them
Hands to a human, immediately
- Password resets, MFA changes and account unlocks. It does not reset, unlock, enroll or elevate anything, for anyone, no matter who they say they are
- Remote access of any kind. It does not log in to a client system, does not run a script, and holds no credentials that would let it
- The diagnosis. It captures symptoms, scope and what still works; an engineer decides what is actually broken
- Anything that looks like a security incident — it reads your client's standing containment instruction word for word and pages your on-call and security lead at the same time
- Out-of-scope or after-hours billable work. It captures the request and routes it for approval by an authorized contact under that client's agreement
- Any caller who asks for a person, on the first ask, without an argument
CallSphere is a front-of-desk agent, not a technician and not a security tool. It does not diagnose a fault, it does not remediate one, and nothing it says is a technical recommendation. On a suspected security incident it reads your client's own written instruction and escalates; it does not decide what happened. And because the help desk is where most social engineering against an MSP starts, the agent is built with no ability to change access at all — it cannot reset a password, remove MFA, or elevate an account, even for a caller who passes every identity check.
What we can and cannot claim
What you can count in seven days
CallSphere is early, and early pilots are what we have. We are not going to hand you a case study with someone else's logo on it and a resolution rate nobody can check. What we will do is put the agent on one client's after-hours line for a week, with every recording and transcript in your hands, and let you count the four things below yourself. The only figure on this page that isn't ours is attributed and linked, and it is about the threat, not about us.
40-60%
of Tier-1 tickets resolved without a human
Illustrative based on early CallSphere pilots; not a guarantee. Actual deflection depends on your ticket mix, knowledge base, and automation rules.
What you can count in seven days
- Every after-hours call answered, recorded and transcribed, so you audit what it said rather than what it reported
- Tickets the agent opened, side by side with the ones your techs opened — same board, same fields, same client
- Escalations timestamped: what triggered the page, who acknowledged it, and how long that took
- A list of the questions it could not answer, so week two is better than week one
Not our numbers — theirs
- 76%
of the ransomware deployments Mandiant investigated were executed outside the victim's working hours — on a weekend, or before 8 a.m. or after 6 p.m. on a weekday
Access, tenancy and recordings
An MSP is a supply chain. So is whoever answers its phone
Your clients pass their audits partly because of you. Putting an agent on your line adds a party to every one of those relationships, so here is exactly what it can reach and what it cannot.
- No client access
- The agent holds no credentials to any client system. It creates and updates tickets in your PSA through a scoped API key you issue and can revoke in one click. It cannot reach an endpoint, a directory, a mailbox or an RMM console.
- Cannot change access
- It cannot reset a password, remove or re-enroll MFA, add a user to a group, or elevate an account. This is not a permission you toggle off — the capability does not exist. The most reliable way into an MSP's clients is still a convincing voice asking the help desk for exactly that.
- Tenant separation
- Every contact is matched to one client before anything is written. An agent configured for one of your clients cannot read or write another client's tickets, contacts, runbooks or recordings.
- Identity checks
- Callers are checked against the authorized-contact list you keep for that client, using the rules that client sets. A caller who fails still gets a ticket — it is routed for a human callback instead of acted on, and the failure is on the record.
- Call recording
- Recordings and transcripts are encrypted in transit and at rest, retained for the window you set, and deleted when it ends. Recording law differs by state and your clients' users are in a lot of states, so the agent reads the disclosure you configure at the start of the call.
- Who can see what
- Role-based access on the console, so a Tier 1 login is not an admin login. Every action is written to a timestamped log: who, what, when, and against which ticket.
- Your clients' regimes
- If your clients carry HIPAA, PCI DSS, CJIS or CMMC obligations, those flow to you and then to us. Tell us which ones before the pilot rather than after. Some change what we can retain and for how long, and one of them may mean we are not the right fit for that client yet.
There is no certificate that makes an agent safe on an MSP's phone line. What there is: a scoped key you control, a capability set that stops short of anything that changes access, and a log of everything it did. Ask any vendor for those three before you ask about a badge.
Questions buyers actually ask
Before you put it on the main line
Can CallSphere handle Tier 1 IT support tickets?
Yes. The AI resolves common issues like password resets, VPN setup, and printer troubleshooting, then escalates unresolved issues to your team with full context.
Does it integrate with our ticketing system?
CallSphere integrates with ConnectWise, Autotask, Freshdesk, Zendesk, and other PSA/ticketing platforms to create, update, and close tickets automatically.
How does the AI triage support requests?
It asks diagnostic questions, checks against your knowledge base, and assigns priority levels before routing to the right technician or queue.
Can it support multiple clients for MSPs?
Absolutely. CallSphere supports multi-tenant configurations so each client gets a customized experience with their own greetings, SLAs, and escalation rules.
What's the average resolution rate for AI-handled tickets?
Our IT support customers see 40-60% of Tier 1 tickets resolved without human intervention, freeing engineers for complex work.
Does it reset passwords or unlock accounts?
No. It cannot reset a password, remove MFA, unlock an account or elevate anyone, and there is no setting that turns that on. Help desks are the most reliable route into an MSP's clients, so the safest agent is one with no ability to change access at all. It verifies the caller against your authorized-contact list, opens the ticket with everything the engineer needs, and routes it.
How does it know which of our clients is calling?
By the number first — inbound caller ID matched against the contacts in your PSA — and by asking when there is no match. Each client can also have its own number or extension. Nothing is written until the caller is matched to one tenant, and an agent configured for one client cannot see another client's tickets, contacts or runbooks.
What actually happens on a P1 at 2 a.m.?
It applies that client's priority matrix, opens the ticket, and pages through the on-call tool you already use, whether that is PagerDuty, a phone tree or a rotation in a spreadsheet. It writes the page time and the acknowledgment back to the ticket. If nobody acknowledges inside the window you set, it walks your escalation path to the next person on it.
Can it set priority, or will it get that wrong?
It sets priority from the matrix you give it for each client: users affected, service affected, whether a workaround exists. It can raise a priority when the caller gives it new information, and it never lowers one on its own — a downgrade waits for a human. Every classification is on the recording, so you can check its judgment in week one instead of trusting it.
What does it do about tickets that are waiting on the client?
It chases them. On the schedule you set and inside the contact hours each client specifies, it calls, texts or emails the authorized contact, asks the one question the ticket is blocked on, and moves the ticket out of the waiting status when the answer comes back. Those are usually the oldest tickets on your board and the ones your SLA reports quietly exclude.
Do we need a new number, or can it sit on our existing line?
Either. Point your current after-hours number at it, or forward only outside your covered hours and leave the daytime line exactly as it is. Most MSPs start with overflow — it picks up when the queue is deep or nobody answers in three rings — and move it to the whole night once they have read a week of transcripts.
How long before it is on our line, and how much work is the second client?
Twenty-four hours to go live on one client with your greeting, that client's priority matrix and your escalation path. The pilot runs seven days. Adding clients after that is configuration, not a new build: what differs between them is the matrix, the covered hours, the authorized contacts and who gets paged.
Put it on one client's after-hours line for seven days
One client, one number, one week. Your greeting, your priority matrix, your escalation path, live in 24 hours. You keep every recording and every transcript, and you will know by Thursday whether the tickets it opens are ones your engineers would have had to rewrite. After the pilot it is $149/mo, or $50/mo if all you want is the line answered and the status questions handled.
Seven-day pilot · One client, one number · You keep the recordings
Overnight summary
Wed · 7:55 AM
- Calls and messages answered
- 7
- Tickets opened
- 4
- Taken off hold
- 1
- Scheduled to the on-call engineer
- 3
- Escalated as P1
- 1
- Sent to voicemail
- 0
Illustrative · one evening