Get started

Hania Assistant

Every workspace comes with an assistant you can talk to. Tell it what you want in plain words and it does the work in your account: building an agent, connecting a tool, adding knowledge, running a test, or tracking down why something isn't behaving.

It is the fastest way to get something done in Hania, and a good way to learn the product. If you would rather do a job by hand, every screen it touches is still there for you to use.

Where to find it

Open Assistant at the top of the dashboard sidebar. That starts a fresh conversation with it. You can also pick it when you start a new conversation, and earlier conversations with it stay in your Conversations list like any other.

The assistant is part of your workspace and only ever sees your workspace. It isn't listed with your own agents, it doesn't count towards the agent limit on your plan, and its settings are managed by Hania rather than edited by you.

Guided setup

When your workspace has something specific to set up, because you signed up for it or picked it on Get started, the assistant already knows what it is. Choose Set it up with Hania and it opens a conversation that takes you through the whole job a step at a time, checking each step works before it moves on. You can leave and come back: the same conversation is waiting, and it knows what is already done.

While it works you see one short line saying what it is doing, such as looking something up or checking your site, with a clock beside it. When it finishes, the line shows what was done and how long it took. Click the line to see the details behind any step.

What you can ask for

Anything you would otherwise click through the dashboard to do:

Build an agent. Describe the job in a sentence or two and it creates the agent, writes the prompt, and picks sensible settings. Ask it to change the tone, tighten the rules, or switch a voice and it edits what it already made.
Connect a tool. Ask for the integration you want and it sets it up, asks you for the credential when one is needed, then tests the connection so you know it works before you rely on it.
Work out an unfamiliar service. If you name a service Hania has no ready-made integration for, it can read that service's documentation, try the calls, and set up a working tool from what it learns.
Add knowledge. Point it at a page or paste in the text you want an agent to answer from.
Put an agent to work. Get it live on your website, on a phone number, or on a messaging channel.
Run an agent for you. Ask it to run one of your Autonomous Agents and it starts the run in the background, so you can keep talking while the agent works. When the run finishes, fails or is stopped, a note appears in the conversation with a link to the run, and the assistant tells you how it went. If the run stops to wait for you, the note comes once it is done. If you pressed Stop in the chat, the note is still added but the assistant doesn't pick it up on its own. Chat Agents can't be run this way.
Build a team. Ask it to put a few of your agents together under a lead and it sets up the team. It can start a team run for you too.
Keep track of work. For list-like work, such as posts to write or leads to follow up, it can set up a collection, add and find items, share it with your agents and write the rules that hand items to them or ask you for a review. Rules it writes start off until you turn them on.
Test and fix. Ask it to try one of your agents and it talks to that agent the way a customer would, in a separate test conversation, then tells you what the agent said. A Tested card in the chat lets you follow the agent's steps as it works and open the test conversation itself. A test doesn't count as a customer conversation. If the agent stops to ask you something or needs your approval for a step, the request waits in the Review queue and the assistant waits with it until you answer. It can also look at a conversation that went wrong or explain why a tool call failed, then apply the fix.
Check your account. Your plan, your balance, what you have used, how many agents you have, what ran recently.
Tidy up. Turn something off, remove an agent you no longer use, rename things so the list makes sense again.

How to ask well

Plain language is enough. Two habits make the results better:

Say what you want, not how to do it. "Make a support agent for my Shopify store that can look up orders" gives it everything it needs.
Be clear about scope. A wide request gets wide work. If you only want one thing done, say so: "just create the Airtable connection, don't touch anything else."

A single request can take the assistant several steps. It works through them and then reports back once, so a short wait with no reply is normal on bigger jobs. You can watch what it is doing as it goes, and open any step to see the details.

Changing your mind mid-job is fine. Press Stop to halt what it is doing, then tell it what you want instead.

When it asks a question

Sometimes the assistant needs to know something before it goes on: which of two agents you meant, or which tool to connect. It asks in the conversation. A card appears with one or more questions, each with choices to pick from or a box to type in. Pick an answer, type your own instead, or leave a question for the assistant to decide. You can also just keep typing in the chat; that dismisses the card and the assistant carries on from your message.

Once you answer, it picks up the same reply where it paused. Your answers stay on the card rather than appearing as a new message from you.

When it asks permission

The assistant stops and asks before anything that costs money or cannot be undone. Buying a phone number, deleting an agent, or deleting a team or taking an agent out of one, for example. You get a card showing exactly what it intends to do, and three ways to answer:

Approve and it carries on.
Approve with edits if the details are nearly right; correct them first and it uses your version.
Decline, with a reason if you want to give one. The reason matters. Tell it "I don't have access to that account" and it will suggest another way round instead of guessing.

If you have stepped away, the request waits for you in the Review queue under Operate, with anything else that needs a decision, such as approvals from collection rules. The assistant's requests stay open there for 72 hours.

When it needs a credential

Some tools need an API key or a token before they can do anything. The assistant asks for it with a secure form rather than in the conversation, so the value goes straight into encrypted storage and is never part of the chat transcript. Once it is saved, the assistant tests the connection and tells you whether it worked.

Some apps don't need a key at all. For those the card has a Connect button: click it, approve on the app's own page in the new tab, and come back to the conversation. The assistant carries on by itself once you have approved. WhatsApp, Facebook Pages and Instagram work a little differently: you sign in with Facebook in a pop-up window and pick your number, Page or account. For WhatsApp you then click Done on the card. If you can't do it right now, say so on the card and the assistant moves on with whatever doesn't depend on it. See Connecting with a button.

Nobody can read a stored credential back afterwards, including the assistant. If a key stops working or you rotate it, the assistant will ask for the new one the same way.

Give each tool the narrowest credential that will do the job, a read-only key where reading is all you need. That limit holds at the service itself, which is stronger than any setting in Hania.

What it will not do

It doesn't act on its own. There is no schedule and nothing to trigger. It works when you talk to it and then stops. The one exception is a run it started for you: when that run ends, it comes back to report on it, up to three times for each message you send. Any later results are noted in the chat, and it sees them the next time you write. For work that should happen on a timetable, build a scheduled agent and let the assistant set it up for you.
It isn't a customer-facing agent. It can't be put on your website, a phone number, or a messaging channel. Those are for the agents you build.
It cannot be deleted or duplicated. It is a fixed part of every workspace, and nothing outside your workspace can be given a way to talk to it.

Things worth asking

"Set up a receptionist for my clinic that books appointments and answers questions about opening hours."
"Connect my Google Calendar and check it works."
"Ask my support agent about our refund policy as if you were a customer, and tell me what it says."
"My WhatsApp agent stopped replying yesterday afternoon. What happened?"
"Add the returns policy on my website to the support agent's knowledge."
"Which of my agents cost the most last month?"
"Test the sales agent as if I were a customer asking about pricing, and tell me how it did."