Combine several automations on one WhatsApp number
Several automations can share one WhatsApp Business number because only one answers an incoming message. They are ranked by match: keyword automations first, then First message, then After hours, then the catch-all, with the oldest winning within the same kind. Design the set from that order: exact answers on keywords, a greeting for new chats, a night reply and an AI for the rest.
easyCoexistence runs as many automations as you need on the WhatsApp Business number you keep in the app, picks one per message in a fixed order, and pauses the set on a chat you answer from the phone.
Automations that answer one incoming message, however many match it.
The priority order, most specific first; within one kind, the oldest automation answers.
On automations per number, with no extra charge and no cap on runs.
The pause after a phone reply is set per automation; the one picked for a message decides whether it is paused.
Runs kept in each automation's Runs tab, which shows which automation took a message and why it stopped.
How does a number decide which automation answers?
In four moves, the same for every incoming message. First, only automations that are switched On and not paused by a refused AI key take part. Second, they are sorted by match kind: Has keyword, then First message, then After hours, then the catch-all called Every message in the editor. Third, within a kind, the automation created first comes first. Fourth, the first one whose trigger fits the message is picked, and only then do its own guards run: the pause after a phone reply, the 24-hour window and the hourly AI cap. If a guard stops it, the run is Skipped and the message gets no automatic reply.
What does a good set of automations look like?
Four layers, one per match kind, each doing the job it is best at. Exact facts go on keywords, so prices and payment details never depend on a model. The greeting goes on First message, with an AI step after it so the newcomer's question is answered in the same run. Nights go on After hours. The AI catch-all takes the rest. Give every layer the same pause hours under Advanced, so the whole set steps back together when you reply from the phone.
Which automation answers which message?
Run a few real cases through the order before you switch the set on. With the four layers above, this is who replies. The cases that surprise people are the last three: a newcomer's keyword beats the welcome, a newcomer at night gets the welcome rather than the night reply, and a paused chat gets nothing from any layer.
- Regular customer, 14:00, "Do you have oat milk?": the catch-all AI.
- Regular customer, 23:00, same question: the After hours automation.
- Anyone, any hour, "what's the pix key?": the keyword automation with the fixed key.
- New customer, first message is "price?": the keyword automation. The welcome never fires for that chat, because the chat is now known.
- New customer at 23:00 writing "hi": the welcome, since First message ranks above After hours.
- A chat you answered from the phone an hour ago: the picked automation is paused there, so no reply.
Why does a skipped run not pass the message to another automation?
Because the pick happens before the guards. Once an automation is chosen, its pause, the 24-hour window or the AI cap can stop it, and the message is then done: the next layer is not tried. That is deliberate, since a pause should silence the chat, not hand it to another automation. Two consequences matter when you design a set. Pause hours should match across layers: a keyword layer with 0 answers in the middle of your conversation while the AI layer with 4 stays quiet. And a layer that is switched Off, or paused by a refused AI key, does drop out before the pick, so the next match answers in its place; to retire a layer for a while, switch it Off rather than editing its trigger.
How do waits and AI limits behave across a set?
Both reach across layers in ways a single automation does not show. A run parked on a Wait is replaced only when a new message picks the same automation; if a keyword layer answers the customer in between, the parked follow-up in the AI layer still goes out later, unless you replied from the phone. So keep follow-ups in the layer that catches most replies, usually the catch-all. The hourly AI cap counts AI replies per chat across the number's automations: the night AI and the catch-all AI draw from one count, and each checks it against its own "AI replies per chat per hour, at most" value. Set that value the same on every AI layer to keep the limit predictable.
How do you test a set before customers see it?
In three passes. In each keyword automation, type samples in the Test it box: it answers "Runs: it matches" with the keyword, or "Does not run: no keyword matches". Then Run test each automation to read its replies; a test runs one automation's steps on the sample, without checking its trigger or the order. Finally, check the order itself from a second phone: write as a new customer, then with a keyword, and open each automation's Runs tab to see which one took each message. To change the order within one kind, recreate the automation that should lose, so it becomes the newest; names do not affect the order.
Can your AI build the whole set at once?
Yes. With the easyCoexistence connector in ChatGPT or Claude, describe the four layers in one request. A careful assistant first calls list_automations to see what already runs on the number, so it does not create a second keyword layer that shadows yours, then create_automation once per layer and test_automation on a sample for each. Ask it to create the layers in priority order and to use the same takeoverHours on all of them.
You: On my number, set up: fixed replies for "price" and "pix",
a welcome that greets new customers and then lets the AI answer,
an AI reply after hours (Mon to Sat, 8:00 to 18:00, São Paulo),
and the AI for everything else. Same 4-hour pause on all.
→ list_numbers {}
→ list_automations { "connection_id": "…" }
→ create_automation { …"match": "keyword", "keywords": ["price","preço","preco"]… }
→ create_automation { …"match": "keyword", "keywords": ["pix"]… }
→ create_automation { …"match": "first_message", steps: reply_text, ai_reply… }
→ create_automation { …"match": "outside_hours", steps: ai_reply… }
→ create_automation { …"match": "any", steps: ai_reply… }
→ test_automation { "automation_id": "…", "text": "quanto custa?" } × 5Common mistakes
- Giving two keyword automations the same word. The older one answers it for good, and the newer one looks broken; keep each word in one automation.
- Expecting the welcome to answer the newcomer's question. First message takes that message, so the catch-all AI never sees it; put Reply with AI after the greeting.
- Setting a different pause on each layer. A chat you are handling then hears from the layer with the shortest pause; use one value across the set.
- Keeping the app's greeting or away message on beside the set. They fire from the phone outside this order, and customers get two replies.
Build the layers in the editor or ask ChatGPT or Claude through the connector; AI layers run on your own key. US$ 9 down to US$ 0.75 per number per month, falling as you add numbers, with the first 3 days free.
easyCoexistence is a multichannel API and MCP server that connects AI agents such as Claude and ChatGPT to WhatsApp, Instagram, TikTok, Facebook and more. On WhatsApp it uses Coexistence: the business keeps using its number in the WhatsApp Business app while the AI works through Meta's official API.
Frequently asked questions
Can two automations reply to the same message on purpose?
No. One message, one automation. When you want a fixed text and an AI answer together, put both steps in the same automation, Reply with text first and Reply with AI second.
Does switching an automation off and on change its place in the order?
No. While it is Off it does not take part, and the next match answers. Switched back On, it keeps its original creation date, so it returns to the same place within its kind.
Do automations on different numbers affect each other?
No. Each number has its own set, its own order and its own chats. What they share is the account's AI keys, one per provider, so a refused key pauses the AI layers on every number that uses that provider.
Can each layer use a different AI model?
Yes. Each Reply with AI step names its provider and model, so the night layer can run a small, fast model and the catch-all a larger one. Each provider bills its tokens to your own key.
Keep reading
Ready to get started?
Set up WhatsApp Coexistence in minutes, not months. The app keeps working on the phone.
Start Free TrialVerified on