Automation
Automation whose "wait 24 hours" still works after the server restarts
Most rule engines run in memory. A deploy, a crash, a scale-down, and the follow-up that was supposed to fire tomorrow never fires — silently, with nothing in the log. Oneop persists every pause as a database row, leases it before executing, and only marks it consumed after the side effects have actually happened.
No credit card. Free plan, 2 seats.
Persisted pending row
Crash between Leased and Done: the row is left runnable, and the next tick re-runs it from step 4 — nothing lost, nothing double-fired.
Three things that make this engine trustworthy
An automation you cannot test and cannot trust to survive a deploy is not automation, it is a hope with a UI.
Delays are durable, and the resume is crash-safe
Delay hit
Writes a pending row: resume time, step index, variables, trace, a snapshot of the flow itself.Tick leases it
Every minute, due rows get their resume time pushed 5 minutes out before running.Step executes
The row is marked done only after execution returns — never before.Crash mid-run?
The row stays runnable. The next tick re-runs it. Nothing lost, nothing double-fires.
A dry-run simulator that writes nothing
Run any flow against a sample ticket for the exact trace — which conditions held, which branch fired, what each AI step produced, what each action would have changed. Actions evaluate for real; only the database write is suppressed, so a mistyped parameter shows up as a no-op before you arm the rule.
Approval steps that wait for a named seniority
An approval step pauses the flow and files a pending row naming the role required to release it, checked against a role rank — an agent cannot release a step that needs a manager. A forbidden attempt does not burn the approval. Reject, and the flow stops there.
The step vocabulary
A Oneop automation rule is a trigger plus an ordered list of steps, and there are six kinds of step.
| Step kind | What it does |
|---|---|
| Condition | A gate. Clauses matched with all or any; if it fails the flow ends as skipped. |
| Branch | Evaluates each branch's clauses in order and runs the first matching branch's inline steps, then continues. |
| Action | Assign an agent, set priority, set status, add a tag, escalate, send a notification, or post a public or internal auto-reply. |
| Delay | Pauses for an amount of minutes, hours or days — optionally counted in business hours only. Persists and resumes. |
| AI | Sentiment, classify, summarize or draft. A metered Gemini call with an optional confidence floor. |
| Approval | Pauses until a human of sufficient role rank approves or rejects. |
Triggers
Ticket created, updated, status changed, priority set, agent assigned, closed — plus SLA breach, described below.
Condition operators
Equals, not equals, in, contains, and the four comparisons — against ticket fields or variables from earlier steps.
Assignment strategies
Name an agent, round-robin, or the least-loaded agent computed from live open-ticket counts.
Bounded by design
48 steps, 8 AI calls per run, shared across nested branches — no branch tree spends your credits in a loop.
Testing is a product surface, not a checkbox
Oneop treats "will this rule do what I think" as a question the product should answer before it runs in production.
A saved rule or an unsaved draft, run against a sample ticket, returns a step-by-step trace: what happened, and how long each step took.
AI steps produce a deterministic signal without billing tokens — the same variable shape the real call produces, so downstream conditions evaluate faithfully instead of falling through an empty context.
Pauses are surfaced, not skipped over — a delay reports "would pause here" and the trace continues, so the whole flow shows in one pass.
Ten templates that are real flows
Every template in the library installs as a real draft rule you can open, read and edit — not a description of a rule.
Installing one creates a draft. You read the steps, change what you want, and arm it when you are ready.
| Template | What it does |
|---|---|
| Intelligent skill routing | Classify the ticket, tag it with the topic, assign the least-loaded agent |
| NLP triage and tagging | Classify, tag and set priority on arrival |
| SLA breach guard | Act on tickets approaching or past their deadline |
| Sentiment escalation | Read sentiment, escalate the unhappy ones |
| SLA breach escalation ladder | Step a breached ticket up successive escalation levels |
| KB deflection | Match a knowledge-base article and reply with it |
| Post-resolution CSAT | Trigger the satisfaction survey after resolution |
| Auto follow-up and close | Chase, wait, then close if nothing comes back |
| Agent copilot | Summarize and draft for the assigned agent |
| CRM context sync | Pull the requester's CRM context onto the ticket |
AI inside a flow, metered and gated
An AI step in a Oneop flow is a real Gemini call, charged to your workspace, with a confidence floor you set.
Four operations ship: sentiment (−1 to 1, plus a label), classify (category, priority, up to three tags), summarize and draft. Strict JSON out, only the ticket's subject and first 2,000 characters in.
The confidence gate is the important part. Below the floor you set, a step returns "below threshold — no action" rather than guessing. A failed call degrades to "AI step unavailable" and the run continues.
Every call passes the same metered choke point as the rest of Oneop's AI — it charges credits and honours your cap. At the cap: a refusal, not an overage invoice. More on AI agents and AI governance.
Escalation, notifications and anomaly watch
Automation is only useful if the right person finds out, and Oneop routes notifications by permission rather than blasting every admin.
Escalation sweep
Every five minutes, breached and unresolved tickets are raised a level, reassigned to the rule's named manager, and both the prior and new owner are notified. Idempotent — running it twice never escalates twice.
One notification service
Recipients resolve by permission and per-category preference, repeats inside 60 seconds collapse into one counted row, and critical events bypass mutes. WebSocket push plus a 60-second polling fallback, instant email for critical items, and a 90-day auto-purge.
Anomaly watch
Every fifteen minutes, each rule's last hour of real runs is compared against its prior six-hour baseline — flagging a rule that has started firing a hundred times an hour.
Opt-in Sending SLA breaches into your automation flows is off until you ask for it. While it is off, breaches are handled by the built-in escalation ladder above. Switched on, breaches fire your flows instead. It is deliberately one responder or the other, so a ticket is never acted on twice.
What it costs
Support automation is on by default on every plan; what changes by plan is the AI budget it can spend and the SLA policies it can act on.
Extra seats are $5 per seat per month. Annual billing is ten months' price for twelve. SLA policies sit on a Pro-and-above plan limit. AI steps draw on the same credit pool as the rest of the platform, and credits are weighted by the work — a chatbot turn and an agent action do not cost the same. Business-plan workspaces can bring their own Google AI key.
See the full comparison| Plan | Per month | Seats included | AI credits/month |
|---|---|---|---|
| Free | $0 | 2 | 50 |
| Starter | $19 | 3 | 1,000 |
| Pro | $59 | 10 | 5,000 |
| Business | $129 | 25 | 15,000 |
Frequently asked
What happens to a scheduled delay if the server restarts mid-flow?
Nothing is lost. The pause is a database row holding the resume time, the step index, the accumulated variables and a snapshot of the flow. A tick leases the row by pushing its resume time five minutes forward before executing, and only marks it done after execution returns. A crash between those two points leaves the row runnable, so a later tick re-runs it from the same step. The trade is at-least-once delivery rather than lost work.
Can I test a rule before it touches real tickets?
Yes. The dry-run simulator takes a saved rule or an unsaved draft plus a sample ticket and returns the full execution trace — conditions, branches, AI outputs, where it would pause and what each action would change — while writing nothing to the database. AI steps produce a deterministic signal in the simulator without billing tokens, so downstream branches evaluate faithfully.
Is the AI in automation real, or a rules engine with a badge?
Real. Sentiment, classify, summarize and draft steps make Google Gemini calls through the same metered choke point as every other AI action in Oneop, charged to your workspace with a hard cap instead of overage billing. Each step can carry a confidence floor, and a model that is not confident enough returns "no action" instead of guessing. Conditions, branches and actions are deterministic logic, and we do not describe those as AI.
Which triggers can start an automation?
Ticket created, ticket updated, status changed, priority set, agent assigned and ticket closed. SLA breach is available as a seventh trigger once you switch SLA-breach automation on; it is off until you ask, and while it is off breaches are handled by the built-in escalation ladder instead.
Can a human approve a step before it runs?
Yes. An approval step pauses the flow and records the role required to release it. The approver’s role is checked against a rank, so an agent cannot release a step that requires a manager, and a rejected attempt does not consume the pending approval. Approve and the flow resumes from the next step; reject and it stops.
Build the rule. Dry-run it. Then arm it.
Start on the free plan with two seats. No credit card, no sales call, and every claim on this page traces to code you can ask us about.
Sign-up opens when the app launches. We will email you once — no marketing list.