Creating your assistant

An assistant is one working personality with one job: the front desk that answers your phone, the salesperson who handles enquiries on WhatsApp, the one that calls people back a week later. Each has its own instructions, its own channels and its own test button.

You can create them yourself, from Bots in the dashboard. Nobody at Kooday has to be involved, and nothing you create can reach a customer by accident — a new assistant starts as a draft, switched off on every channel, and stays that way until you deliberately bring it live.

This page covers both halves: making a new one, and living with one you already have.

Where to find it: Bots, under Assistants in the left-hand menu. The page header reads “Select a bot to see its configuration and test it live in your browser — the same pipeline a real call runs through.”

The five-minute version

1. Click Create assistant (top right).
2. Pick a kind, type a name, click Create.
3. Click Open studio, write what it should do in the Instructions pane, click Save draft, then Publish.
4. It comes live by itself. Back on Bots, open the Readiness tab and click Validate setup to confirm.

That is a working assistant. Switching it on for real customers is the next section down, and it is a separate decision on purpose.

How many you can have

Your plan sets the number. Click Create assistant past that number and the dialog refuses with “Your plan’s assistant limit is reached” and a line telling you what you have:

Your plan includes 3 assistants and all are in use. Contact your administrator to add more.

Two things about that count:

• Drafts count, archived ones don’t. The number includes every assistant on the account that isn’t archived, live or not — an abandoned draft still occupies a slot, so tidy up before you assume you are capped. Archiving an assistant you no longer need frees its slot immediately; deleting it is not required.
• Only Kooday can raise it. There is no self-serve upgrade in the dialog — the refusal says “Contact your administrator” because that is genuinely the only route.

Step 1 — Create it

Create assistant opens a dialog that says exactly what it will do:

Pick what kind of assistant this is and give it a name. It starts as a draft — nothing goes live until you write and publish its instructions.

There are only two things to fill in.

The kind

Four to choose from, and the choice sets its default manner — how it opens, how hard it pushes, what it does when it cannot help:

Kind	Suits
Service desk	The front desk. Someone arrived with a specific need; answer it quickly and exactly.
Sales	Enquiries you want to convert. Direct, curious, confident — and not desperate.
Mentor	Teaching and guidance. Warm, unhurried, supportive rather than transactional.
Outbound follow-up	Reaching people who did not ask you to get in touch. Brief and deliberately unpushy.

Pick by the job, not by the department. A clinic’s reception line is a Service desk even though bookings are revenue.

You cannot change the kind later from the dashboard. It is not fatal — your own instructions override the manner in practice — but if you are torn between Service desk and Sales, decide now rather than after you have written a page of instructions.

The name

Type what your team will call it — Front Desk, Reception, Sales. Up to 80 characters. It is used verbatim in the assistant’s own starting instructions, so a name a customer could plausibly hear works better than an internal code.

Kooday derives an internal id from the name behind the scenes; you never see or choose it. Two consequences worth knowing:

• The name needs at least one a–z letter or digit. A name written only in Malayalam or Devanagari script is refused with “The assistant’s name must contain at least one latin letter or digit (a–z, 0–9) — it is used to build its internal id.” Give it a latin name and let the instructions set the language it actually speaks.
• Two assistants can share a name. Nothing stops you, and the second quietly gets a numbered id — so the picker shows Sales — sales and Sales — sales2, which nobody on your team can interpret. Give them different names.

Click Create. You land on the new assistant with a message: “Front Desk created — write its instructions to bring it live.” In the selector at the top it appears as Front Desk — frontdesk (draft).

Step 2 — Write its instructions, and publish

From the assistant’s card, click Open studio to reach Prompt Studio — its dedicated screen. The Instructions pane is where a draft becomes an assistant. It carries the subtitle “How this assistant should behave, on every channel — identity, greeting style, do’s and don’ts.”

Who this assistant is

Above the editor is a card with two fields — its own name and how it opens — and they belong to this assistant alone, not to your business. Full walkthrough of both fields: Prompt Studio → Who this assistant is.

Both fields save as a draft, like everything else here, and reach customers when you Publish. They are compiled into the instructions below, in a block marked as generated — your own text is never touched, and editing inside that block does not stick.

There are two buttons, and the difference is the whole point.

Button	What it does
Save draft	Stores your text. Changes nothing a customer hears.
Publish	Makes it what the assistant actually says, from that moment.

The line under them says it plainly: “Saving never changes what your assistant says — Publish does.” Above them you get the current state — Nothing published yet, or Live: v3 · unpublished draft.

Publishing is what brings a new assistant to life

There is no “activate” button. Publishing the instructions on a draft is the activation. Kooday re-runs the setup check the moment you publish, and if it passes, the assistant flips from draft to live on its own.

If the check does not pass, your text is still published — but the assistant stays a draft until the problems are fixed. The Readiness tab back on Bots tells you what they are.

Write behaviour, not a script

Instructions describe how it works: who it is, what it may promise, what it must never do, what to do when it does not know. Facts belong in your knowledge base and prices belong in your catalogue — an instruction block stuffed with prices goes stale the week you change them.

⚠️ A typo in a {{TOKEN}} blocks the publish. If you write something that looks like a fill-in slot Kooday does not recognise, the save warns “Draft saved, but publish will refuse it” and lists the offending tokens. Fix the typo, or ask us to set that token up.

You can undo. History lists every version, Restore brings an older published one back, and Unpublish stops the assistant using it at all — with the warning “Unpublish this content? Your assistant stops using it immediately; drafts are kept and you can publish again later.”

Step 3 — The setup check

The Readiness tab holds one card, Setup check, and one button, Validate setup. It “Verifies the bot’s prompt templates and tool bindings exist and are wired correctly — the same checks the platform runs before a call.”

You do not have to remember to run it — it re-runs on every change and at restart, and the tab shows the last result with Last checked underneath. Click it when you want a fresh answer right now.

Four states, and only one of them needs you:

What you see	What it means
Not validated yet — click “Validate setup” to check this bot.	It has never been checked. Click it.
✓ Setup looks good — no problems found.	Nothing to do.
No blocking errors, but 2 warnings below.	It will work. The warnings are worth reading once.
3 errors — this bot will fail its calls until fixed.	It is broken and calls will fail.

The badge beside the assistant reads Ready, Has issues or Not checked to match.

Most errors here are platform wiring rather than anything you typed. Two are yours to fix, and both name themselves plainly:

• “…is not a language this assistant can speak — it would answer in English instead.” The language on the assistant is not one we serve (often a typo). Pick a supported one. Leaving the language as mixed is fine and is never flagged — that means “follow the customer”.
• “…is not a voice ‹provider› offers any more.” A warning, not an error: calls still go through in the default voice. Pick a current voice when convenient.

Most of what shows up here is yours to fix. Occasionally the problem is on our side — a shared building block the platform maintains, not anything on your assistant. When that happens the panel says so plainly (never a raw technical message) and offers a Report to Kooday button that sends us the details in one click — no need to copy and paste the wording yourself.

Step 4 — Channels, and the surprise

The Channels card on the Configuration tab is where an assistant reaches actual customers. Five rows:

Channel	What switching it on does
Voice	Answer phone calls and in-browser voice widgets.
WhatsApp	Reply automatically to inbound WhatsApp messages.
Instagram	Reply automatically to inbound Instagram messages.
Messenger	Reply automatically to inbound Facebook Messenger messages.
Web chat	Serve the website chat widget.

⚠️ A brand-new assistant has every one of these switched off — and most of them locked. This is deliberate, and it is the opposite of what people expect from a wizard that just said “created”. The locked message tells you what to do next, and it is different depending on the channel:

• WhatsApp, Instagram, Messenger show “Connect through Meta sign-in to enable” with a link to Integrations — this is genuinely something you can do yourself, no administrator needed. See Connecting a channel for the Meta sign-in walkthrough.
• Voice shows a greyed-out box reading “Contact your administrator to enable this channel” — a phone number is still something Kooday sets up with you.

Web chat is the one you may be able to switch on yourself with no lock at all, and only if your plan includes it.

That is the honest shape of self-serve creation today: you can build an assistant end to end, and connect it to WhatsApp/Instagram/Messenger yourself through Meta sign-in — a phone number is still something Kooday does with you.

Two rules that catch people out:

• You cannot switch a channel on while it is still a draft. Publish the instructions first. Channel changes on a draft are rejected.
• Switching a channel off always works, immediately, whatever your plan says. Existing conversations stay in your Inbox; the assistant simply stops answering new ones. There is no “pause” that is gentler than this and none is needed.

On, but not actually on

WhatsApp has one more state, and the card says so: “On, but not yet active — no WhatsApp connector selected below.” The switch is on, but no WhatsApp account is bound to this assistant, so nothing reaches it. Pick a connector further down the same tab.

Instagram and Messenger have the same trap without the warning. Switching either on with nothing wired behind it looks successful and does nothing. Verify with a real message before you believe it.

Answering voice notes

Once WhatsApp is on, a small extra choice appears: “When someone sends a voice note, reply with” — Text (Transcribe the voice note and answer in text) or Voice note (Answer with a spoken voice note). Text is the safer default; voice notes feel personal but cost more and cannot be skim-read.

Step 5 — Test it before anyone else does

The Live test tab appears once Voice or Web chat is switched on, and holds whichever applies.

Test this bot places a real call from your browser: “Place a call from your browser. Allow microphone access when prompted — you’ll hear the bot through your speakers and it hears you.” Click Start call, talk to it, Hang up. Tick Record this call and you get a Download recording afterwards — the file stays in your browser. It runs the real call pipeline, so what you hear is what a customer hears.

Test the chat bot is the typed equivalent: “the same brain that answers WhatsApp and web chat. The bot follows the conversation, so you can test a multi-step flow; Clear starts a fresh one. Nothing here is saved.”

⚠️ The chat tester has a tick box that makes it real. “Run this bot’s real tools” is off by default, and the help text is not being dramatic: “When on, this test fires the bot’s real webhooks, bookings, and CRM writes against your live systems — and the turn is recorded in the audit log.” Leave it off unless you specifically mean to fire a real booking.

If the tab says “Live tester not available”, the in-browser tester is not switched on for your server — tell us and we will enable it.

Living with an assistant you already have

Everything above except the create dialog applies to an assistant Kooday built for you. A few things only matter once one is running.

The Configuration tab, field by field

Most of what an assistant answers with is yours to change directly, inline, on the Configuration tab — click a value and it saves the moment you leave the field (or flip the switch, for a toggle). There is no separate Save button per field.

Field	What it controls
Name	What your team calls it in the dashboard. Not what it calls itself on a call — that is the How it opens field on Prompt Studio.
Phone number	The number this assistant answers on, in E.164. Only shown once Voice is on. See Connect your Exotel account for how it is validated against a connected credential.
Language	A language name (e.g. malayalam), or mixed to follow the customer.
Default inbound language	The language an inbound call starts in, before the customer says anything. Voice only.
Max call length (seconds)	The hard cap on how long a single call may run. Voice only; must be a positive number.
Recording	On or off — a consent decision, not a technical one. Decide it against how you tell callers you record, not against whether the transcripts are useful.
Knowledge search	Whether this assistant may search your knowledge base mid-conversation. Off is a real, supported state — some assistants should only use what you told them directly.
Voice engine, provider and voice	Which speech engine this assistant runs on, and — for the cascade engine — which speech-to-text, language model and text-to-speech provider it uses. See below.
Persona instructions	Free-text notes layered into this assistant’s prompt, on top of its bound persona. Optional.
Tools	Read-only. The list of tools an assistant can call is set by Kooday, not from here.

A plain team member (not an admin) sees every value on this list, but cannot edit any of them — the row renders as text rather than a control. Only tenant_admin (and Kooday staff) can change a bot’s configuration.

Voice engine and providers

Every voice-capable assistant runs on one engine:

Engine	What it is
Gemini Live (streaming)	One model handles listening, thinking and speaking together.
Cascade (STT → LLM → TTS)	Three separate steps you can mix and match — pick your own speech-to-text, language model and text-to-speech provider, each independently, or leave any of them on Default.
OpenAI Realtime (streaming)	A fused streaming engine, with its own model and reasoning-effort pickers.
Azure Realtime (streaming)	Same shape as OpenAI Realtime, on Azure’s models.

The two Realtime engines need a plan that includes premium voice. If your plan does not, they simply do not appear in the Engine dropdown. If a Realtime engine was set on your assistant before your plan changed, the dropdown still shows it — labelled “(current — not on your plan)” and disabled — so the picker never lies about which engine is actually answering calls, even though you cannot re-select it.

Switching the Engine changes which of the other fields appear: Cascade shows a speech-to-text/language-model/text-to-speech provider picker plus a voice; the two Realtime engines show a voice, a model and a reasoning-effort picker; Gemini Live shows only a voice. Default on any provider or voice field means “use the platform’s current default for this engine” — it is a real, recommended choice, not a placeholder you need to fill in.

Voice changes apply to the assistant’s next call, not one already in progress.

Notifications wires the WhatsApp message that goes to you after a call finishes. It appears only for assistants that take calls, and it is a completely different thing from the WhatsApp channel — one sends you a summary, the other lets an assistant answer your customers. They were one switch once and it was a mistake.

You cannot deactivate or delete an assistant yourself. Switching every channel off is the self-serve equivalent and it is complete — the assistant answers nobody. Ask us for the rest.

Troubleshooting

The Create assistant button does nothing, or errors

Creating assistants needs admin access. The button is shown to everyone but only admins can use it.

I published my instructions and it is still a draft

The setup check did not pass. Open Readiness, click Validate setup, and read the errors. Publishing only activates an assistant that is otherwise ready.

There is no Live test tab

It appears only when Voice or Web chat is on. A draft has neither, so publish the instructions and switch a channel on first.

The channel switch is greyed out

That channel has not been granted to this assistant — the row says “Contact your administrator to enable this channel.” It is not something a setting on your side can fix.

I switched WhatsApp on and nothing happens

Look at the row again. If it reads “On, but not yet active — no WhatsApp connector selected below”, the switch is on but no WhatsApp account is bound to this assistant. Instagram and Messenger fail the same way without saying so.

I changed something on a draft and it did not save

Draft assistants reject configuration changes. Publish the instructions to bring it live, then change the setting.

Where to go next

• Connecting a channel — getting it onto WhatsApp, your website, or a phone number.
• Your knowledge base — the documents it answers from.
• Selling from your catalogue — so it quotes real prices.
• Setting up your sales pipeline — turning the good conversations into deals.