> ## Documentation Index > Fetch the complete documentation index at: https://docs.quivly.ai/llms.txt > Use this file to discover all available pages before exploring further. # Agents > Build automated workflows on a visual canvas — triggered by record changes, schedules, Slack messages, or health score changes — with AI steps, branching, and human review gates. Agents are automated workflows. Each agent has a trigger, a series of steps on a visual canvas, and a run history. Think of an agent as a playbook that runs itself: "when a customer's health drops, gather context, draft an outreach, and ask me to approve it." ## Creating an agent From the **Agents** page you can build manually or describe what you want in plain language — for example, *"When an opportunity closes won, create an onboarding project and notify the CSM on Slack."* The AI builder picks the trigger, selects tools, drafts instructions, and wires the canvas. It may ask one clarifying question, and it warns you if a similar agent already exists. You land on the canvas with an AI chat panel open — keep describing changes and accept or reject each proposed edit. ## Triggers | Trigger | Fires when | | -------------- | ------------------------------------------------------------------------------- | | Record Created | A new record is created in the selected object (with optional field conditions) | | Record Updated | An existing record changes | | Schedule | On a cadence (daily, weekly, quarterly, or every N days) for a customer segment | | Slack Message | A message is posted in a mapped customer channel | | Health Score | A customer's health bucket changes (optionally to a specific bucket) | Record triggers support filter conditions (equals, contains, greater than, is empty, and so on) and enrollment control — run once per record, or re-enroll every time. ## Steps Steps come from a searchable palette, grouped into: * **AI** — run a [skill](/product/skills) with structured input and output. * **Decision** — If/Else branching, Switch routing by field value, and **Review**: pause until a human approves (approvals show up in [Actions](/product/actions)). * **Actions** — send a Slack message or DM, create an [AI notebook](/product/notebooks), create or update projects, tasks, and milestones, create/update/delete records, send a webhook, or run an **App Action** in a connected app like Gmail or Google Calendar. * **Lookups** — read customer data (health scores, calls, revenue, usage, contacts, tickets, and more) to feed later steps. For user-scoped apps (Gmail, Calendly), you choose who the action sends as: the record's CSM owner, a specific teammate, or whoever published the agent. Deleting records requires a Review gate. ## Template agents The trigger, step, lookup, and action primitives combine into playbooks for most customer-facing workflows. These five are the ones teams reach for first — each is buildable today from the palette above, either by hand or by describing it to the AI builder. | Agent | How it's built | What it does | | ----------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Churn save play** | Health Score trigger (bucket drops) → Lookups (health, calls, usage) → AI [skill](/product/skills) to draft outreach → Review → Slack DM to the CSM | Catches a health decline the moment it happens, packages the context, and puts a ready-to-send message in front of the owner for approval. See [Identifying at-risk customers](/workflows/identifying-at-risk-customers). | | **Onboarding kickoff** | Record Updated (opportunity → Closed Won) → create onboarding project with tasks and milestones → notify the CSM on Slack → App Action to send a welcome email | Turns a won deal into a fully scaffolded onboarding plan and a first-touch email, with no manual setup. | | **Renewal prep brief** | Schedule (N days before renewal) → Lookups (revenue, contracts, usage, health) → AI [notebook](/product/notebooks) for a renewal brief → Review → assign a prep task | Gives the owner a data-backed renewal brief on a predictable cadence, ahead of every renewal date. | | **Slack signal triage** | Slack Message trigger (mapped customer channel) → AI skill to classify and summarize → If/Else branch → create an [Action](/product/actions) and DM the owner on risk, or log a note otherwise | Reads every customer channel message, routes the urgent ones to a human, and quietly files the rest. | | **Expansion spotter** | Schedule or Record Updated (usage crosses a threshold) → Lookups (usage, contacts) → AI skill to spot the opportunity → Review → draft outreach via App Action | Surfaces accounts ready to grow and drafts the expansion outreach for review. | **Use cases these unlock:** churn prevention and account saves, hands-off onboarding, renewal and QBR prep, revenue expansion, support escalation, and scheduled executive or segment reporting — anywhere a repeatable "when X happens, gather context and act" playbook applies. ## Publishing and runs Agents are drafts until you publish; each publish creates a version. You can test-run from the builder before going live, and pause an agent by toggling its status. **Run History** shows every run with its status (Pending, Running, Awaiting Review, Completed, No Action Taken, Failed, Cancelled), version, whether it fired automatically or manually, and a full step-by-step trace. Running or paused runs can be cancelled. ## FAQ Only if you build it that way. Add a Review step before any outbound action and the run pauses until someone approves it in the Actions inbox. Each customer in the segment runs once per cadence window — a daily agent won't run twice for the same customer in a day. No. Agents are built on a visual canvas, and the AI builder can assemble one from a plain-language description.