AI customer support with Aamu.app: Helpdesk Knowledge Base, drafts, and human review

Short answer: How Aamu.app uses the Helpdesk Knowledge Base, AI-generated drafts, and human review to make customer support faster without losing control.

AI customer support is most useful when it does not try to replace the whole support team. The practical win is more grounded and less theatrical: every incoming question gets understood faster, every reply starts from the right company knowledge, and the human support person keeps final control over what is sent. The useful version of AI support is less “robot takes over” and more “robot finds the draft while the human keeps the keys.”

Aamu.app is built around that kind of workflow. Support conversations, email threads, tasks, docs, project knowledge, and the team itself live in the same workspace. That matters because AI support is only as good as the context it can use. A model can write fluent text from a vague prompt, but a support team needs answers that reflect the company's actual policies, current work, product details, and previous decisions.

This article walks through how Aamu.app approaches AI customer support with four connected ideas: Helpdesk Knowledge Base for shared knowledge, AI-generated drafts for fast response work, Live chat automation for real-time conversations, and human review before anything sensitive is sent.

Why AI support often disappoints

Many AI support demos look impressive because the question is simple and the answer is already obvious. A customer asks a common billing question, a bot finds a policy paragraph, and the reply looks polished. Real support is messier.

Real support questions often depend on details that live in different places. The answer may require a public help article, an internal operating rule, a recent product change, a known bug, the customer's earlier conversation, and a task that someone in the team is already working on. If an AI tool only sees a disconnected FAQ, it will either answer too generally or invent confidence it has not earned.

There are also organizational problems. If AI replies are sent automatically, the team may lose visibility into what customers were told. If generated answers are not connected to the rest of the workspace, they create another place to check. If the knowledge base is stale, support automation becomes a faster way to spread outdated information.

Aamu takes a different starting point: AI should help the team use its own knowledge better. It should draft, summarize, suggest, and connect work. It should not quietly become a separate support system with its own memory and its own version of the truth.

The Aamu model: knowledge, drafting, live chat, review

Aamu.app separates AI customer support into four layers.

1. Helpdesk Knowledge Base: the shared knowledge layer

Helpdesk Knowledge Base retrieves context from the team’s configured sources, including Docs, web pages and curated answers, before a support reply is drafted.

2. Drafts: AI writes into the normal workflow

When Aamu generates a support reply, the result is a draft. The draft is saved into the same place where the support person would write a reply manually. It is visible, editable, and connected to the ticket or email thread.

3. Live chat: AI can answer in real time when allowed

Drafts are the safest default for email and Helpdesk tickets, but customer support also includes live conversations. Aamu Live chat is part of the same support system, and it can use Helpdesk Knowledge Base as context when AI is allowed to answer live chat messages automatically.

This is controlled by a project setting in the Live chat configuration: Allow AI to answer live chat messages. In the project settings, it lives under the Helpdesk Live chat settings, for example at /project/edit/PROJECT_ID/helpdesk/livechat. When the setting is enabled, the live chat widget can become an AI-assisted first line of support instead of only a way to wait for a human agent.

The key difference is immediacy. A ticket draft waits for a person to review and send it. A live chat answer may be shown directly to the visitor during the conversation, so the knowledge and escalation rules matter even more. The AI should answer questions that are safe, grounded, and covered by Helpdesk Knowledge Base, and it should avoid pretending to know things that require a person.

4. Human review: sending is deliberate

Generating a draft and sending a reply are separate actions. A person can check the facts, edit the text and decide when to send.

What Helpdesk Knowledge Base changes

Without a knowledge layer, a support AI usually depends on whatever context is manually placed into the prompt. That works for experiments, but it does not scale well. Someone has to decide what to include each time. Important details are easy to miss. The prompt grows long, fragile, and hard to maintain.

Helpdesk Knowledge Base changes the shape of the problem. Instead of asking the model to answer from memory, Aamu can first retrieve relevant knowledge and use that context while drafting. The result is a workflow that is closer to how an experienced support person works: look up the right policy, check the latest product detail, consider the customer's specific situation, then write the answer.

This helps in several common support scenarios.

Policy questions

Customers often ask about cancellation terms, refunds, data retention, billing periods, account ownership, or service limits. These answers should not be improvised. Helpdesk Knowledge Base can give the draft generator the relevant policy language so the reply is aligned with the company's rules.

Product usage questions

Support teams answer many questions that are really product education: how to invite a teammate, how to create a task, how to connect email, how to use docs, or how to build an API integration. When the answer can be grounded in product documentation, AI can produce a clearer first draft and link the support reply back to the right concepts.

Known issues and current work

The hardest support questions are often about things that are in motion. A bug has been identified, a migration is underway, a feature changed recently, or a workaround exists but is not yet public documentation. If this knowledge is captured in the workspace, Helpdesk Knowledge Base can help the support draft reflect the team's current understanding.

Tone and operating principles

Good support is not only correct. It also sounds like the company. Helpdesk Knowledge Base can include instructions and examples that shape tone: concise or detailed, formal or friendly, direct or more explanatory. This is especially useful when several people answer customers and the team wants replies to feel consistent.

Language is a separate question from tone. AI can usually answer in almost any language, so a good default is to reply in the same language the customer used unless the team has a reason to do otherwise.

A practical ticket workflow

Here is a typical AI-assisted support flow in Aamu.app.

A customer sends a support message.
The ticket appears in Aamu Helpdesk.
Aamu identifies the ticket as unanswered.
Helpdesk Knowledge Base retrieves relevant knowledge for the question.
Aamu generates a reply draft using the ticket and retrieved context.
The draft appears in the Helpdesk reply editor.
A human reviews the draft, edits it if needed, and sends it.
If the ticket reveals missing knowledge, the team can update docs or Helpdesk Knowledge Base for next time.

Example: cancellation billing question

Imagine a customer writes:

Hi, if we cancel today, are we charged again next month? We are on annual billing and I am not sure what happens to the remaining period.

Suppose the company’s documented policy says cancellation stops renewal but preserves access through the paid period. A draft based on that policy might read:

Hi, thanks for checking. If you cancel today, your subscription will not renew at the end of the current annual period. You can keep using the workspace until the paid period ends. We do not automatically charge another annual period after cancellation. If you want, I can also confirm the exact renewal date for your account.

The support person can then verify the account, add the specific renewal date, and send the reply. The AI has removed the blank page and surfaced the right policy, but the human still handles the final customer-specific detail.

Example: product how-to question

A customer asks:

Can we use Aamu to draft replies for incoming email threads, but make sure nobody sends them automatically?

This is exactly the kind of question where the answer should be product-specific. In Aamu, email and Helpdesk draft generation follows a human-in-the-loop model. AI can generate a draft, and the draft is stored for the selected actor, but sending is a separate explicit action.

A draft answer can explain that clearly:

Yes. Aamu can generate reply drafts for email threads and Helpdesk tickets without sending them automatically. The draft appears in the normal reply editor, where your team can review, edit, and send it. Sending is deliberately separate from draft generation, so you can use AI to prepare replies while keeping human approval in the workflow.

Live chat: automatic answers with Helpdesk Knowledge Base

Live chat changes the timing of support. In email and ticket queues, a draft can wait for a human review. In live chat, the customer is often on the website right now, deciding whether to continue, buy, configure, or ask for help. A slow answer can lose the moment.

Enable Allow AI to answer live chat messages in the project’s Helpdesk Live chat settings to let AI answer visitors directly.

This is where Helpdesk Knowledge Base becomes especially important. A live chat answer should not be a generic model response. It should be grounded in the same company knowledge as the rest of the support workflow: product documentation, pricing rules, setup instructions, known limitations, tone guidelines, and escalation rules.

A practical live chat flow can look like this:

A visitor opens the Live chat widget on the website.
The visitor asks a product, pricing, setup, or troubleshooting question.
Aamu checks the project Live chat settings and sees that AI is allowed to answer.
Helpdesk Knowledge Base retrieves relevant knowledge for the question.
AI sends a grounded answer in the chat conversation.
If the question is risky, unclear, account-specific, or outside the known material, the conversation can be escalated for a human to handle.
The chat remains part of the Helpdesk history, so the team can review what happened and improve the knowledge base if needed.

This is a different level of automation from draft generation. It can be very useful for common website questions: how the product works, whether a feature exists, where to find a setting, how to start a trial, how pricing works, or how to solve a simple setup problem. It is less appropriate for anything that requires account-specific judgment, legal commitments, refunds, security decisions, or emotionally sensitive handling.

Where the Aamu API fits

Aamu also exposes this model through the API. That is useful when a team wants to build its own automation around Aamu, connect another system, or run scheduled workflows.

For example, an integration can retrieve unanswered Helpdesk tickets, ask Aamu to generate reply drafts, and leave those drafts ready for review. Another workflow can query Helpdesk Knowledge Base directly, use the returned context in a custom model prompt, and then create a task or doc based on the result.

The important design principle stays the same: generated content can be prepared through the API, but sending a reply is a separate explicit step.

Retrieving Helpdesk Knowledge Base context

POST /api/v1/team-brain/retrieve
x-api-key: YOUR_API_KEY
x-project-id: YOUR_PROJECT_ID
Content-Type: application/json

{
  "query": "How should we answer a billing cancellation question?",
  "limit": 8
}

The retrieve endpoint returns matching knowledge entries and source chunks. An integration can use those results as grounded context before it asks a model to write anything.

Generating a Helpdesk reply draft

POST /api/v1/helpdesk/tickets/TICKET_ID/reply-draft/generate
x-api-key: YOUR_API_KEY
x-project-id: YOUR_PROJECT_ID
x-aamu-actor: ai
Content-Type: application/json

{
  "instructions": "Answer in the same language the customer used, friendly and concise.",
  "use_team_brain": true,
  "mode": "replace"
}

The generated text is saved as a reply draft in the Helpdesk ticket for the selected actor. It appears where the team reviews and sends ticket replies, and it is not sent automatically.

Sending only when ready

POST /api/v1/helpdesk/tickets/TICKET_ID/reply-draft/send
x-api-key: YOUR_API_KEY
x-project-id: YOUR_PROJECT_ID
x-aamu-actor: ai

How to prepare your knowledge for AI support

The best AI support workflows start before the first draft is generated. A team should decide what knowledge the AI is allowed to rely on and how that knowledge should be maintained.

Write policies in answerable form

AI retrieval works best when knowledge is explicit. Instead of storing only a vague note such as billing handled case by case, write the actual rule, the exceptions, and what support should ask when details are missing.

Capture recurring support answers

If the team answers the same question repeatedly, that answer belongs in shared knowledge. Aamu can help turn repeated replies into docs, Helpdesk Knowledge Base entries, or internal guidance that future drafts can use.

Keep product changes close to support

When a feature changes, support should not have to discover the change from customers. If docs, tasks, and support live near each other, the team has a better chance of keeping AI context fresh.

Define tone and escalation rules

Some questions should not receive a normal AI-drafted answer. Security incidents, legal threats, refund disputes, abuse reports, and emotionally sensitive messages may need escalation. Put those rules into the workspace so the team and the AI workflow can follow them.

What good AI support metrics look like

AI support should not be judged only by how many messages it sends. Faster sending is not automatically better. A better set of metrics looks at speed, quality, and control together.

How much faster can the team produce a first reviewed reply?
How often do drafts require only light editing?
Which questions still need manual research?
Which draft failures reveal missing knowledge?
Are customers getting more consistent answers?
Are sensitive tickets being escalated instead of over-automated?

A simple adoption plan

A team does not need to automate everything at once. A good rollout can be gradual.

Start with one support queue or one type of recurring question.
Add the relevant policies and product notes to Helpdesk Knowledge Base.
Generate drafts, but require human review for every reply.
Track which drafts save time and which ones need major edits.
If using Live chat automation, enable AI answers only after the safe question categories and escalation rules are clear.
Improve the source knowledge where drafts or live chat answers are weak.
Expand to more ticket categories once the workflow feels reliable.