Picture a Tuesday afternoon on a HaloPSA service desk. A user calls about a shared mailbox that stopped syncing. Your tech spends twenty minutes on the phone, rebuilds the Outlook profile, re-adds the mailbox permission in Exchange, and confirms the fix with the user. Then the next call rings. The action that lands on the ticket reads "Called user, fixed Outlook." with 0.25 hours on it.
Now turn on every AI feature HaloPSA has. Halo can summarize that ticket, score its sentiment, suggest a category, match it to similar tickets and draft a knowledge article from it. All of that output will be built from one line of text, because that line is all the ticket holds.
That is the practical question behind "HaloPSA AI": not whether Halo has AI (it does, and a lot of it), but what it reads, and what never makes it into the ticket for it to read.
What is HaloPSA AI?
HaloPSA AI is the set of AI features built into Halo's platform, which Halo markets as Halo AI. It covers ticket summaries, AI suggestions for category, priority and routing, ticket and article matching, sentiment and satisfaction insights, knowledge base article creation, AI acknowledgement emails, and AI agents that work inside Halo's role-based permissions. It runs on a Halo-hosted OpenAI connection by default, or on your own Azure OpenAI, OpenAI or Amazon Bedrock account. It works on the records already in Halo: tickets, their actions and emails, and knowledge base articles.
What HaloPSA's built-in AI does
Halo documents its AI in Leveraging AI in Halo, Connecting AI to Halo and its Halo AI page. Grouped by what an MSP service desk would use them for:
| Feature | What it does | What it reads |
|---|---|---|
| AI Insights | Summary, tonality, sentiment, satisfaction, suggested priority, resolution and type | The ticket and its actions |
| AI Suggestions | Suggests category, priority, agent, estimate, team routing and SLA-breach warnings | The ticket |
| AI category suggestions | Suggests category values, considering up to 50 of them (Q4 2025 release notes) | The ticket's details |
| AI ticket matching | Finds similar tickets and can link matches as child or related tickets | Embeddings built from ticket data |
| Article suggestion and creation | Suggests KB articles on a ticket; drafts articles from ticket correspondence | Tickets and the knowledge base |
| AI acknowledgement emails | Replaces the templated acknowledgement with an AI-written reply per ticket type, using a custom prompt | The new ticket |
| Thank-you detection | Spots "thanks, all sorted" emails on closed or pending-closure tickets | Incoming email |
| Improve agent responses | Fixes spelling, grammar and tone in a tech's note | The note being written |
| AI agents | Conversational agents (web chat, Teams, Slack, portal) and process agents such as dispatch, first-line support and major incident | Halo records, within the permissions you set |
Halo also lists an AI report builder, AI surveys and AI search, and says its native MCP servers let assistants such as Claude, Copilot and Codex work with Halo data.
For an MSP already on Halo, that is a strong starting point. Categorization, routing suggestions and similar-ticket matching are the jobs where most desks feel pain, and Halo covers them without another vendor in the loop.
How AI is connected to HaloPSA
Settings live under Configuration > AI > General Settings. You pick one of these connections:
- Default Halo connection. Halo says it uses the OpenAI API "covered under an enterprise agreement with OpenAI," needs no credentials, and that data "is not stored by OpenAI or used to improve OpenAI." On v2.244 and later the default model is gpt-5.4-nano; earlier versions use 4o-mini.
- Your own Azure OpenAI. You enter the endpoint, API key, API version, GPT deployment name and an embeddings deployment (1536 dimensions). See Connect AI to Halo using Azure OpenAI.
- Your own OpenAI account, billed by OpenAI.
- Your own Amazon Bedrock (v2.244+). Halo notes that not all AI functionality is available on Bedrock.
Two things to check before you switch features on. First, not every feature runs on every connection: Halo's guide says some, including AI article creation and improving agent responses, need your own provider connection rather than the default one. Second, you can mark a customer Exclude from AI functionality when a contract or regulator says their ticket data can't go to a model.
On cost, Halo's AI page says its AI agents are "included across all plans" with no separate AI licence, and the default connection is described as free to use. If you bring your own Azure, OpenAI or Bedrock account, you pay that provider for usage. Confirm the details for your plan with Halo.
The limit: Halo's AI reads the ticket, not the call
Every feature in that table works on text that already exists in Halo. None of Halo's AI pages describe transcribing phone calls or reading call recordings.
On an MSP desk, that matters more than it sounds. A large share of the actual work happens in places Halo never sees as text:
- The phone call. The troubleshooting, the user's answers, the fix, and the time are in the conversation. The ticket gets whatever the tech types afterward.
- The remote session. Twenty minutes inside a user's machine or a client's Microsoft 365 admin center turns into one sentence, or nothing.
- The back-and-forth before work starts. A ticket that says "printer broken" needs three questions answered before anyone can work it. Someone has to ask them.
So the quality ceiling on Halo's summaries, sentiment scores, matched tickets and generated KB articles is the quality of the notes your techs write. A model can't summarize steps nobody recorded, and a ticket-matching engine can't connect two tickets whose fixes were never written down. This is the same gap we covered for Autotask AI and ConnectWise AI: the AI inside a PSA is downstream of documentation, not a replacement for it.
The billing side has the same shape. If a 20-minute call is logged as 15 minutes, or not logged at all, no AI insight on the ticket will recover it. We went through where that time leaks in MSP time tracking.
Where a capture layer fits next to Halo AI
The useful split is: Halo's AI works the record; something else has to fill the record. That is the job Vight does on HaloPSA. Vight is listed in Halo's integrations directory, and here is what it writes into Halo, specifically:
Call summaries as Halo actions. Vight pulls recorded calls from your phone system (3CX, Microsoft Teams, RingCentral, Zoom Phone, Dialpad and others), transcribes them, matches each call to the ticket it was about, and posts a summary of the issue, the steps taken and the resolution as an action. Matching uses ticket numbers spoken on the call and, with phone lookup on, the caller's recent tickets. When no ticket fits, the call stays unlinked instead of landing on the wrong ticket. More on how summaries are built in AI call summary.
Time on the action, with a charge rate. Set Post as to Time Entry and the call duration becomes the action's Time Taken. Pick a charge rate: Halo logs time posted without one as No Charge, even on a ticket with a contract. When several techs were on the call, each mapped tech gets their own action with the same time. Posting as each technician needs the admin permission on the API application; without it, the summary posts as the integration's agent.
Outcomes that label, not trigger. Actions post with the outcome you choose, Private Note for internal notes by default, and can be hidden from the end user. The outcome only labels the action. Its status changes, emails and other steps you configured in Halo don't run, which is what you want from an automated note.
Triage that asks the questions. Vight's ticket triage runs on HaloPSA. It matches a new ticket to a scenario you configure, emails the requester for the details they left out, posts an internal note with what's answered and what to confirm, and can move the ticket to a ready-to-work status. It learns which questions your techs ask from closed tickets on the Halo teams you pick. It doesn't assign tickets, change priority or close them. Our ticket triage guide covers the process side.
Work sessions, in beta. On HaloPSA, Vight's Work Notes Agent (a Windows desktop agent) turns a technician's work session on a ticket into a step-by-step note and a time entry. Password managers are ignored by default.
A tab inside the Halo ticket. The Vight ticket tab shows a ticket's calls and summaries in the Halo ticket screen, with Ask Vight chat for that ticket. It follows each agent's team access in Halo, so a tech can't use it to read tickets on a team they can't reach.
The order matters. Once calls and work sessions are landing as detailed actions, Halo's own summaries, matching, category suggestions and KB article drafts have real material to work with.
HaloPSA permissions Vight needs
If you're evaluating any tool that writes into Halo, ask exactly which API permissions it needs. For Vight, the API application uses Client ID and Secret (Services) authentication, under Configuration > Integrations > HaloPSA API:
- Required:
read:tickets,edit:tickets,read:customers. - Optional:
read:assets(devices in ticket briefs),read:calendar(client follow-up),edit:customers(client pop-ups),admin(one-click ticket tab install, per-technician attribution and automatic work note posting),read:timesheets(automatic work note posting).
A new Halo API application starts with no permissions, so every request fails until you grant them. That is the most common first-day setup issue.
How to decide what to turn on
A sequence that works for most Halo shops:
- Turn on Halo's AI Suggestions and category suggestions first. They act on every ticket, cost nothing extra on the default connection, and show you how good your ticket text is.
- Read twenty AI summaries on phone-heavy tickets. If they say "user called, issue resolved," your bottleneck is capture, not AI.
- Fix capture on calls before you lean on KB generation. Articles drafted from thin tickets become thin articles.
- Check time and charge rates. Compare a week of call durations from your phone system against the time on matching Halo actions. The gap is your unbilled time.
- Decide on data boundaries. Use Halo's customer exclusion for clients who can't have data sent to a model, and ask any add-on vendor how it handles the same clients.
FAQ
Does HaloPSA have AI built in?
Yes. Halo includes AI suggestions (category, priority, routing), AI insights (summary, sentiment, satisfaction), ticket and article matching, KB article creation, AI acknowledgement emails, thank-you detection and AI agents. It's configured under Configuration > AI.
Is HaloPSA AI an extra cost?
Halo's AI page says AI agents are included across all plans with no separate AI licence, and the default Halo connection is described as free to use. If you connect your own Azure OpenAI, OpenAI or Bedrock account, that provider bills you for usage.
What model does HaloPSA AI use?
By default, Halo's own OpenAI connection. Halo's guide says v2.244 and later use gpt-5.4-nano and earlier versions use 4o-mini. You can connect your own Azure OpenAI, OpenAI or Amazon Bedrock instead.
Can HaloPSA AI transcribe or summarize phone calls?
Halo's AI documentation does not describe call transcription or reading call recordings. Its features work on text already in Halo. To get calls into tickets as notes, you need a tool that records and transcribes the call and posts the result to Halo as an action.
Can AI post time entries in HaloPSA?
Yes, through an integration that writes actions with Time Taken. In Vight, a call summary posted as a Time Entry carries the call's duration and the charge rate you choose; time posted without a charge rate is logged as No Charge.
Does AI triage work on HaloPSA?
Halo's AI Suggestions suggest category, priority and routing. Vight's triage agent also runs on HaloPSA and goes a step further on intake: it emails the requester for missing details and notes the answers on the ticket.
Where to start
If your Halo AI summaries look thin, the fix is upstream of the model. See how calls become HaloPSA actions with time and the right charge rate on our phone to PSA page, or book a demo and we'll walk through it on a Halo ticket.