Question 1

Should I paste my whole brand book in here?

Accepted Answer

No. Facts belong in Knowledge. Instructions are a handful of sharp rules about tone, format, and what to refuse.

Question 2

Can an instruction teach the chatbot a fact?

Accepted Answer

Not reliably. The chatbot answers from your knowledge base, and an instruction only steers wording. Put the fact in Knowledge instead.

Question 3

Do instructions run Call actions?

Accepted Answer

No. Instructions shape wording and policy only. Actions live on the chatbot's Actions tab and are triggered by their own description.

Question 4

Why is the chatbot ignoring my rule?

Accepted Answer

Usually it was not saved, it is too vague to act on, or your knowledge says the opposite. Check Save first, then make the rule concrete.

Question 5

Are instructions shared between my chatbots?

Accepted Answer

No. Each chatbot keeps its own list, and each list is capped separately by your plan.

Question 6

Does a Viewer use up an instruction slot?

Accepted Answer

No. Slots belong to the chatbot, not to teammates. A Viewer simply cannot save changes to the list.

Question 7

Can I buy extra instruction slots?

Accepted Answer

No. There is no instruction add-on. The count comes from your plan, so changing plan is the only way to raise it.

Question 8

I upgraded and the maximum has not changed.

Accepted Answer

Confirm Billing shows the new plan on the workspace that owns this chatbot, then reload Settings → Instructions.

Question 9

What happens to my instructions if I downgrade?

Accepted Answer

Nothing is deleted. You simply cannot add more until you are back under the smaller cap.

Question 10

Is a Call action the same as a channel?

Accepted Answer

No. Channels are where people talk to your chatbot. Actions are what it may do once they are talking.

Question 11

Can the chatbot invent an action I never created?

Accepted Answer

No. Only actions you created and enabled are offered to it, and it cannot reach anything you did not configure.

Question 12

Can I trigger an action myself from the dashboard?

Accepted Answer

No. Actions run only inside a conversation, at the moment the chatbot decides one fits. Nothing runs on a schedule.

Question 13

Do actions cost extra credits?

Accepted Answer

No. The reply that triggered the action spends message credits at the chatbot's model rate, the same as any other reply.

Question 14

Which type should I start with?

Accepted Answer

A Button. It sends nothing and stores no secret, so it proves the whole pipeline works before you wire up anything risky.

Question 15

Why is Create disabled?

Accepted Answer

You are at your plan's enabled-action cap, you lack permission on this chatbot, or Call actions are switched off for this deployment.

Question 16

Does creating an action cost credits?

Accepted Answer

Saving costs nothing. A test with a sample message makes a real call to the chatbot's testing model, so it needs message credits left to run, but it is not charged as a reply.

Question 17

What is the difference between saving and enabling?

Accepted Answer

A saved action sits in the library and never runs. Only an enabled action is offered to the chatbot during a conversation.

Question 18

Can I import an API I already have?

Accepted Answer

Yes. Paste an OpenAPI or Swagger document as JSON and each operation becomes an action to review and save. YAML is not accepted.

Question 19

Why does my action ask the visitor to confirm?

Accepted Answer

It changes data. Anything using POST, PUT, PATCH or DELETE needs a confirmation on the visitor's next message before it fires.

Question 20

Does a template contain a secret?

Accepted Answer

No. Every template ships with the product and is identical for everyone. You supply your own credential.

Question 21

My service is not in the gallery.

Accepted Answer

Build it from scratch as a Custom API action, or bridge it through the Zapier or Make template.

Question 22

Can I edit a template after saving it?

Accepted Answer

Yes. Once saved it is an ordinary Custom API action, so the address, parameters, body and credential are all editable.

Question 23

Will Agentency update my action if a provider changes their API?

Accepted Answer

No. Template updates do not reach actions you already saved. A provider change shows up as failed runs that you fix yourself.

Question 24

Why is my template asking the visitor to confirm?

Accepted Answer

It creates something. Templates that book, order, ticket or send ship with confirmation switched on.

Question 25

Which templates need no credential at all?

Accepted Answer

The webhook ones: Slack, Discord, Microsoft Teams, Zapier and Make. The webhook URL is itself the secret, so keep it private.

Question 26

Which templates only read data?

Accepted Answer

Product look-ups, order status, shipment tracking and web search. They ship without a confirmation step because nothing is changed.

Question 27

Why is there no template for my CRM?

Accepted Answer

Services needing an OAuth sign-in flow, form-encoded bodies or GraphQL cannot be templated directly. Route them through Zapier or Make.

Question 28

Do I need one template per store?

Accepted Answer

Yes. Each action carries one address and one credential, so two stores means two actions.

Question 29

Why did a template that used to work stop?

Accepted Answer

Usually an expired or rotated credential. Salla merchant tokens in particular expire every 14 days.

Question 30

Can the chatbot change where the request is sent?

Accepted Answer

No. You fix the scheme and host at save time. Values from the conversation fill the path, query and body only.

Question 31

Can I call an API on my own network?

Accepted Answer

No. Private, internal and reserved addresses are refused and logged as Blocked destination. Use a public HTTPS hostname.

Question 32

Is this the same as the MCP connector?

Accepted Answer

No. A Custom API action is a single request you configure yourself. MCP is a separate type that is usually unavailable.

Question 33

What happens if my endpoint is slow?

Accepted Answer

Each call is time-boxed. Past the budget the run is recorded as Timed out and the chatbot tells the visitor it could not do that.

Question 34

Could a retried reply create two orders?

Accepted Answer

No. A repeat of the same change replays the recorded outcome instead of firing the side effect again.

Question 35

Does handoff open a live chat window for my team?

Accepted Answer

No. Agentency has no live-chat inbox for staff. Handoff is an escalation signal with a link to the transcript, so you reply your own way.

Question 36

Where does the alert go if I leave the email blank?

Accepted Answer

To the chatbot owner's account email, so an escalation is never lost because someone left the field empty.

Question 37

What language is the alert written in?

Accepted Answer

The chatbot owner's saved language. The reason the chatbot gives for escalating stays in the visitor's language.

Question 38

Does a handoff stop the chatbot replying?

Accepted Answer

No. The conversation continues normally and the chatbot keeps answering from your knowledge.

Question 39

Where do I see how often handoffs happen?

Accepted Answer

The Handoff Rate KPI on your main dashboard shows the share of conversations escalated in the selected date range.

Question 40

Can a button send a message back into the chat?

Accepted Answer

No. Buttons are links out. If you want tappable questions that send text into the chat, use Suggested messages (Settings → Suggested messages) instead; the only other buttons that reply are the automatic Confirm and Cancel pair on actions needing confirmation.

Question 41

Does Agentency count button clicks?

Accepted Answer

No. Add your own campaign parameters to the URL and read the clicks in your own analytics tool.

Question 42

Can I style the button?

Accepted Answer

Not directly. It follows your chatbot's accent colour, and the text colour is picked automatically so it stays readable.

Question 43

How many buttons can appear at once?

Accepted Answer

As many as genuinely match the question, but each one is a separate enabled action counting against your plan allowance.

Question 44

Why is my URL rejected?

Accepted Answer

It needs a full http:// or https:// address. A bare domain such as example.com is not accepted.

Question 45

How is this different from a handoff?

Accepted Answer

Notify is silent and does not flag the conversation. Handoff tells the visitor someone will follow up and counts toward Handoff Rate.

Question 46

Can the alert go to Slack or Teams instead of email?

Accepted Answer

Not with this type. Use a Custom API action with the Slack, Discord or Microsoft Teams template, which needs only a webhook URL.

Question 47

Who writes the summary in the alert?

Accepted Answer

The chatbot, at the moment it fires. That is what makes the alert readable instead of a bare notification.

Question 48

Why are we getting far too many alerts?

Accepted Answer

The trigger description is too broad. Name the specific signal and exclude questions your knowledge base already answers well.

Question 49

Is this the same as the pre-chat form?

Accepted Answer

No. Lead capture asks before the first reply and can block the chat. Collect info asks mid-conversation, when it is relevant.

Question 50

Where do the collected details go?

Accepted Answer

Into your Customers list, saved on the server. They never travel back to the visitor's browser with the chat reply.

Question 51

What happens if I add no fields at all?

Accepted Answer

Agentency falls back to name as optional and email as required, so a minimal action still works.

Question 52

Will the same person be saved twice?

Accepted Answer

No. A returning visitor who gives the same email is matched to their existing customer record.

Question 53

Can it read a form already on my page?

Accepted Answer

No. It can only save what the visitor actually typed in the conversation.

Question 54

What does Skipped actually mean?

Accepted Answer

A confirmation the visitor never answered. It is not a catch-all, and it counts as neither a success nor a failure.

Question 55

Can I replay a run?

Accepted Answer

No, deliberately: replaying would send a second email or create a second order. Use the Test tab for a dry run that contacts nothing.

Question 56

There are no runs at all. Is that a bug?

Accepted Answer

No. An empty log means the action was never attempted, so check it is enabled and that the description matches how visitors ask.

Question 57

Do sandbox tests appear in the log?

Accepted Answer

No. Test-tab runs send nothing and record nothing, by design.

Question 58

Why is my success rate still low after a fix?

Accepted Answer

It is a lifetime figure, not a rolling window. Filter the log by date to judge how the action is behaving now.

Question 59

How do I tell whether MCP is on?

Accepted Answer

Open the New action chooser. No MCP tile means it is off. If the tile is there, Discover tools is the definitive test.

Question 60

Will upgrading my plan enable MCP?

Accepted Answer

No. It is a deployment-level capability rather than a plan feature, so no upgrade turns it on.

Question 61

Is MCP required to go live?

Accepted Answer

No. Knowledge, a welcome message, and an active widget or hosted link are enough to launch.

Question 62

Will enabling MCP later delete my other actions?

Accepted Answer

No. The action types sit side by side and nothing you already configured is disturbed.

Question 63

What is the closest alternative?

Accepted Answer

A Custom API action, or an OpenAPI import when you want many endpoints exposed at once.

Instructions and actions FAQ