What technicians use internal chat agents for
An Internal chat agent carries your full MSP toolbox, so builds span the whole service desk. Five patterns cover most of them:| Pattern | A technician asks it… | What it draws on |
|---|---|---|
| Research & lookup copilot | ”Summarize ticket 44821”, “find the SOP for re-enabling MFA”, “what did we do last time this client had this?” | Read-only search across your documentation, configurations, and past tickets — the most common build |
| Queue & dispatch | ”Give me my next ticket”, “what’s unassigned right now?”, “triage my open queue” | Reads the live PSA queue and assigns or routes work — see the self-serve dispatch recipe |
| On-demand ticket actions | ”Reply to the customer”, “log 15 minutes and set it to In Progress”, “schedule a service call”, “close it” | Ticket updates, notes, time entries, and service calls — gated behind approval |
| Guided provisioning | ”Onboard these three new starters”, “offboard jdoe@acme.com”, “run the workstation-setup checklist” | Microsoft 365 / AD user, group, and license changes against a runbook — gated behind approval |
| On-demand reporting | ”Which Acme firewalls have warranties expiring in 90 days?”, “top 10 noisy devices this month”, “is Huntress installed on this machine?” | Pulls from PSA, RMM, and documentation, then builds a table or exportable file |
Create the agent
Create
Go to Chat → New Chat Agent (or create an agent in chat mode). There is no Triggers & Filters step — chat agents start when someone messages them.
Set the audience
Pick the agent’s Audience — it’s fixed once set and shapes everything below:
- Internal — for your own technicians. Carries your full MSP toolbox (PSA, RMM, Microsoft 365, cross-company lookups). Runs in the dashboard, as a custom internal Teams bot, and is also what the Neo Support Agent is.
- End-user — for your clients’ employees. A deliberately narrow, company-scoped tool set, served through a branded end-user bot. See how end-user bots work.
Write instructions
Custom instructions define the agent’s job, tone, and limits. Be explicit about what it should refuse and when to ask a technician. If end users will talk to it, write for them — not for your technicians. See Writing effective agent instructions for the full guide.
Enable tools and permissions
Pick tools and set per-integration permissions exactly as for other agents. Enable the minimum the job needs — every enabled write permission is reachable from a chat message.
Settings that work differently in chat
| Setting | Behavior for chat agents |
|---|---|
| Trigger & filters | None — user-initiated |
| Technician-in-the-Loop | Always delivered as an inline approval card in the chat |
| Test mode report | Always reported in the chat reply |
| Model & reasoning | Platform-managed; no per-agent override |
| Mention mode | In Teams group chats and channels: mention-only (default) responds only when @mentioned; read-all responds to every message. 1:1 chats always respond. |
Who can chat with the agent
By default every chat agent is open to your whole team. An admin can change who may message a specific agent from Roles & Access → Chat Agent Access:| Option | Who can message the agent |
|---|---|
| Everyone on your team | Anyone on your team — the default |
| Admins only | Your admins only |
| Specific people | Admins, plus the individual people you choose |
Per-client instructions for end-user agents
If the agent serves end users through a channel, you can add Chat Instructions per client company (Companies → edit company). They’re injected when a user from that company is talking — use them for company-specific policies, escalation contacts, or tone.Before going live
- Require Technician-in-the-Loop on write permissions.
- For end-user-facing agents, prefer read-only permissions plus “create a ticket” over direct writes.
- Test the agent as an end-client at a chosen company before you roll the bot out — you’ll see its per-company instructions and end-user tools exactly as that company’s users will, without installing anything.
- Share a read-only preview with the client before rollout — a public, no-login link that shows what the assistant does, with sensitive values masked.