How a tool ends up “automatic”
Always-available tools
Added for one kind of run
Neo adds these only where they can work: the intent loaders when an intent applies, and the question buttons on chat agents.Neo Support Agent tools
Only the Neo Support Agent has these. No other agent gets them, and you can’t add them to one.Dependency helpers
These are added automatically when an enabled tool needs them — you never select them.Auto-injected integration tools
These turn on when you configure the matching integration’s permissions on the workflow’s Integrations tab. They’rehidden_from_ui (you manage them via the integration, not a tool toggle):
Verify Dispatch Decision
Neo runs one agent per ticket, and those runs happen at the same time. Two runs that read the same technician calendar a few seconds apart both see the same free slot, and both assign it — a double-booked engineer. Before a dispatch agent writes an assignment, this tool reserves that technician and slot for 10 minutes so no other run can take them, re-reads the calendar, and reports anything it finds wrong: leave, an overlapping appointment, a slot outside the technician’s working hours, a slot past the SLA deadline, or a technician who has just been given several tickets. When the agent books a time slot, Neo re-reads that technician’s schedule entries for the slot straight from your PSA, with no cache. It runs last, right before the agent writes, so it can find a booking your run’s earlier calendar reads were too early to see. That matters because the technician who books their own meeting is not a Neo run, and the reservation does not cover them. The report names each clashing entry by its PSA schedule entry id, so your agent can compare it against the entry your PSA quotes if it rejects the write, and can put the id in the ticket note. The reservation stops other Neo runs and nothing else — it books no time in your PSA. Your PSA can still refuse a schedule entry over something created outside Neo, and the receipt tells the agent what to do about it: read the calendar again, pick a slot that clears the entry your PSA named, and write it. One refused slot is a second attempt, not a ticket handed back for manual dispatch. Most assignments set an owner and book no time slot, and on Autotask those get a check of their own: whether the new owner is on leave between now and the end of the next business day. A ticket assigned in the evening reaches its owner on their next working day, which the agent’s earlier calendar read often stopped short of — and the engineer who is off tomorrow has the fewest open tickets today, precisely because they are away. This one is Autotask-only on purpose. Autotask records leave as a time-off request, so “on leave” is unambiguous. On ConnectWise the same calendar flag also covers lunch breaks, end-of-shift markers and all-day onsite visits, so the check would report engineers who are simply out at a customer site as unavailable. The slot checks are unaffected on every PSA — they only report a conflict when your booked slot actually lands on the blocked time. The working-hours check earns its place on a retry. Your PSA checks a new schedule entry for overlap only, so it accepts a booking half an hour after the engineer went home and reports success. When the PSA rejects the first slot for a clash and the agent picks another, that replacement clears the clash and can still land outside the working day. A reported problem does not cancel the assignment. Your agent reads the report and decides, because your custom instructions say what your business accepts — “never book over an existing appointment”, or “a P1 may take the next free hour even if it misses the SLA goal”. You see each decision, and the agent’s reason for it, in Executions. There is nothing to switch on, nothing to configure, and it costs no credits. Two things are worth knowing: the tool cannot force the agent to call it, so write the call into your dispatch instructions; and only the ticket owner is reserved, not any extra resources added alongside them.Analytics category tools
Generate Artifact is always available, but agents don’t file a report unless the work calls for one — instructions that ask for a report or analysis, or a result someone will read later. For routine task execution the agent just writes its dashboard summary. To make an agent always produce a report (or never), say so in its custom instructions.It is also not the only thing that shows up under Artifacts on a run: any file an agent leaves in its sandbox is attached there as a download. Generate Artifact writes a report you read in place; the sandbox produces files you download.
Want fine-grained control over the auto-injected integration tools? You don’t toggle them off — instead set the relevant permission groups to Disabled on the workflow’s Integrations tab. See Tools Overview.