Skip to main content
Neo Agents are the intelligent workers behind your workflows. Unlike simple automation rules, agents understand context, make decisions, and can handle complex multi-step tasks just like a human technician would.
Wondering whether to build an agent or a workflow for your use case? See Workflows vs Agents — start with workflows for triage, dispatch, resolution, escalation, and merge; reach for agents for everything else.

Tools

Discover the tools agents can use to accomplish tasks

Skills

Best practices and guidance agents load on-demand

​
Building Your First Agent

Watch this 5-min walkthrough of building agents in Neo Agent:

​
What Agents Can Do

Agents combine AI understanding with real-world actions. When a ticket comes in, an agent can:
  • Analyze the problem using past tickets, documentation, and web knowledge
  • Communicate with end users to gather information or provide updates
  • Take Action in your PSA, RMM, or Microsoft 365 environment
  • Request Approval from technicians before sensitive operations
Agents can also run on a schedule without a ticket — performing proactive tasks like M365 compliance audits, inactive account cleanup, and recurring admin operations.

Scheduled Agents

Run agents on a recurring schedule for proactive tasks that don’t start from a ticket

Chat Agents

Agents your technicians and end users talk to directly — in the dashboard or via Teams

​
How Agents Work

1

Receive Context

When a workflow triggers, the agent receives the ticket details, company information, and any custom instructions and memory you’ve configured. For scheduled agents, the agent runs from your custom instructions and enabled tools alone — no ticket needed.
2

Analyze & Plan

The agent analyzes the situation, searches for relevant information, and determines the best course of action based on your instructions.
3

Use Tools

The agent selects and uses the appropriate tools to accomplish the task—whether that’s searching documentation, updating a ticket, or communicating with a user.
4

Report Results

After completing its work, the agent provides a detailed log of actions taken, along with any relevant outputs or recommendations.

​
Agent Tools

Tools are the capabilities you enable for your agents. Each tool allows the agent to perform specific actions:
Tools are configured per-workflow. You choose which tools an agent can use and customize their settings to match your processes.

​
Agent Skills

Skills provide specialized guidance that agents load on-demand. Unlike tools (which perform actions), skills teach the agent how to approach a type of task. Skills are always available and loaded automatically when relevant. Custom instructions take priority over skill guidance when there’s a conflict. You can also create your own skills for the repeatable procedures your agents run again and again — a weekly report, a monthly reconciliation — so every run comes out the same reliable way. See custom skills.

Learn more about Skills

See all available skills, including how to build your own

​
Credit Tiers

Tools sit in four tiers: Free (0), Basic (1), Standard (2), Advanced (3). What you pay depends on how the agent starts. The same toolbox is cheap on a ticket-triggered agent and can be expensive on a scheduled or chat agent. External API Request is Advanced (3 credits per call on scheduled and chat). The Neo API tool is Free. The API skills under Built-in Capabilities are also Free to load — they are not the API tools. The full list is on Tool credit costs. To get an email on a day an agent uses far more credits than usual, switch on Daily Credit Alert on the agent’s Overview tab. A daily credit alert for one agent explains how Neo sets its cap.

​
Testing an agent before you trust it

​
From a new agent to a trusted one

Neo asks for your approval before each change the agent makes. What supervision does not cover lists the exceptions.
  1. Run it in Supervised mode on a real ticket. In Executions, choose Run now and turn on Supervised run. The dashboard offers this step as soon as you create an agent.
  2. Approve, reject or comment on each card as it works, in Teams or on the Approvals page.
  3. When the ticket is done, click Train on the run in Executions, say what should change, and send.
  4. Neo Support reviews the run and your decisions and updates the agent when you confirm the edit it proposes.
Repeat on a few tickets. When the runs need no corrections, enable the agent. Supervised mode has what each card shows and what an approval covers.

​
Two questions while you write instructions

Two different questions, and they need different tools. “Does the agent do the right thing with this ticket?” Open the agent in Executions and use Run now. It executes the agent immediately against that ticket, skipping the trigger, so you can change the instructions and run again without waiting for an event. This is the fast loop while you are writing instructions. Run now works on a disabled agent too, and its changes are real unless test mode is on: disabling an agent stops only its triggers and its schedule. To watch it work before you trust it, turn on Supervised run there and approve each change yourself: see supervise a single run. “Does my agent fire when a customer replies?” Run needs a real event, so you have to produce one. Set up a test company with a test contact, raise a ticket for that contact on a board the agent watches, and reply from the contact’s mailbox.
Reply from an address your PSA holds as a contact, not as a technician account. ConnectWise attributes an inbound email to the ConnectWise member whenever the sending address also belongs to one, and a member’s note is a technician note — not a customer reply. So a Customer replied agent does not fire for it, by design.This catches people testing from their own work mailbox, which is usually a member. Your real customers are contacts, so the agent behaves correctly for them; only the test is wrong. Use a mailbox that is not a technician account in your PSA — a test contact on a test company is the simplest setup.Neo records which of these happened. On Executions, the event’s match reason reads no customer note is present, and where Neo can tell which note the reply was it adds : and the cause. Most causes name something in ConnectWise — the member it attributed the note to, or a sender it does not hold as a contact — and those are settings on your side rather than faults. One names Neo’s own reply on the ticket, which flips the same flag ConnectWise uses for a customer reply; an agent does not answer its own note, and there is nothing for you to change. When two notes on the ticket could have been that reply, or the nearest is too far off to be sure, Neo names none rather than guess, so a reason with no cause after it is not an error. If you believe a genuine customer reply is being missed, send us the reason text and the ticket.
Keep test mode on for the first runs either way, so the agent reads without writing.

​
Best Practices

When test mode is on, no action changes your systems: updating tickets, creating time entries, client-facing notes, internal notes and the like are simulated. It still reads, and it still works through the whole job: each write it would make is simulated, and it goes on to the next step as it would on a live run. The writes it would have made appear in the run’s output when the run ends, next to the agent’s run summary — see Report test results via for where else Neo posts that report. Monitor agent decisions closely when first deployed to ensure quality. When test mode looks right, a supervised run lets the agent make real changes, each after a technician approves it, and Train on that run updates the agent from your decisions. See From a new agent to a trusted one.
For sensitive operations, start with Technician-in-the-Loop enabled. This gives you oversight while still benefiting from AI-powered analysis.
The better your instructions, the better the agent performs. Be specific about your processes, preferences, and edge cases. Iterate based on feedback to refine the agent’s behavior. See Writing effective agent instructions for the full guide.