Start from a flow that already works
Every template is the same shape: deterministic steps for what must never vary, and an agent node for the turn a script cannot handle. Open one, point it at your data, and change what you need.
- Ready-made flows
- 6
- Triggers to start a run
- 3
- Actions that do the work
- 8
- Logic blocks, evaluated not inferred
- 3
The flow library
Six flows for the conversations most teams handle first. Pick one to see the canvas, what keeps it inside the lines, and when a person takes over.
Ends with Resolved in one turn, with the live shipment state.
Message received
Any message
Agent
orders.lookup · shipping.track
Matches the customer, then reads the order and its tracking state
If / else
Order found?
Send status
Answer
Ask for order ID
Clarify
- What keeps it safe
- The agent can read an order, but it cannot invent one. If the customer cannot be matched to a record, the flow branches to a clarifying question instead of guessing at an order number.
- When a person takes over
- Two failed identification attempts, or any order flagged as disputed, goes to a human with the full transcript attached.
- Runs on
- WhatsApp, Chat
- Reads from
- Shopify, PostgreSQL, Salesforce
Every block, and what it does
A flow is a trigger, some steps, and the branches between them. Every block below is deterministic and does exactly what it says. The agent node is the one exception, and the only one that needs guardrails.
Triggers
What starts a run. Every flow begins with exactly one.
- Message received
- Any inbound message on a connected channel, whatever the customer wrote.
- Keyword matched
- A specific word or phrase, for the cases you want caught before the agent ever sees them.
- Campaign reply
- A response to something you sent, so the flow already knows what it is answering.
Actions
The steps that actually do something, inside or outside RoutMe.
- Send template
- An approved WhatsApp template, the only way to open a conversation outside the 24-hour window.
- Send text
- A free-form reply inside an open conversation window.
- Add tag
- Label the contact so later flows and segments can find them.
- Set attribute
- Write a value onto the contact record, such as a plan tier or a preferred language.
- Assign agent
- Hand the conversation to a person, with everything said so far already attached.
- Call a URL
- Post to your own endpoint mid-flow when the work belongs in your systems, not ours.
- Wait
- Hold the run for a fixed duration before the next step.
- Delay until time
- Hold until a specific clock time, so a 2am trigger does not become a 2am message.
Logic
Where a run splits. These are evaluated, never inferred.
- If / else
- One condition, two branches. The workhorse of every flow here.
- Switch
- Many branches off a single value, for cases like order state or plan tier.
- Filter
- Stop the run unless a condition holds, which is how you keep flows from firing on the wrong contacts.
Most tools make you choose one
A rule builder you can trust but cannot talk to, or an agent that talks well and cannot be pinned down. The interesting work is in putting them on the same canvas.
Rules only
Predictable, and it breaks the moment a customer phrases something the way people actually phrase things.
- Never goes off-script
- Easy to audit
- Every phrasing needs a branch
- Falls back to a human fast
Agent only
Handles any phrasing, and gives you no way to guarantee the refund window is respected every single time.
- Understands what was meant
- No branch explosion
- Policy left to judgement
- Hard to explain after the fact
Both, on one canvasRoutMe
Deterministic steps hold the rules that must never vary. The agent node takes the open-ended turn in the middle.
- Policy stays a rule
- Agent handles the language
- Every tool call logged
You can see exactly what it did
Every branch taken and every tool called is logged as it happens, not reconstructed afterwards. When a customer gets a strange answer, the trace is already there.
- The trigger fires and the flow starts
- The agent node reads the message and picks its tools
- Each tool call runs against your systems
- The branch after the agent decides what happens next
- Delivery outcomes come back on your webhook
The same flow runs on every channel you enable
Adding a channel is a configuration change, not a rebuild. The flow, the agent, and the data it reads stay exactly as they are.
- WhatsAppLive
- ChatLive
- EmailSoon
- InstagramSoon
- MessengerSoon
- TelegramSoon
- ViberSoon
- WeChatSoon
- LINESoon
- SMSSoon
- RCSSoon
- IVRSoon
Pick a flow. Point it at your data.
Every template here is a starting point, not a demo. Change what you need and put it on a channel.