How to Build an AI Agent Without Code in 2026: From Your Website to a Working Voice and Chat Agent in 5 Steps

In short: In 2026 you can build an AI agent for the phone or your website chat without writing a line of code. The work is not in the technology but in five decisions: what the agent knows, how it behaves, what it may do, when it hands over to a person and how you test it before launch. Start with one channel and one task. Only expand once real conversations show that it works.

We revised this post in October 2026. The previous version promised all channels in every EchoCall plan. It listed additional messaging channels, gave a latency figure, presented telephony as part of the price and promised that the €1 verification amount would be paid back. The claims about phone numbers in more than 60 countries, AI notices in 74 languages and that most agents resolve 70 to 80 per cent of requests after two weeks were not backed up either. All of that is gone. WhatsApp is currently not part of what you set up yourself in the EchoHub. Today the EchoCall chat agent runs on your website, with handover to your team.

What an AI agent must be able to do before you build it

An AI agent takes in a request, decides on the next step based on your knowledge and carries out that step. So it does not just answer, it gets something done: it books an appointment, creates a ticket or puts the caller through to a colleague. A detailed explanation is in What Is an AI Agent?, and the difference from a classic bot is covered in AI Agents vs. Chatbots.

That leads to three requirements for everything you are about to set up:

• Your own knowledge instead of general knowledge. The agent should answer from your content, not from whatever a language model read somewhere.
• Tools. Without access to a calendar, CRM or ticketing system, every conversation ends with: "You'll have to call us, I'm afraid."
• A boundary. What the agent does not know, it should not make up. It needs a route to a person.

In its guide to building agents, OpenAI writes that companies typically achieve greater success with an incremental approach than with a fully autonomous system from day one. The advice is to get the most out of a single agent first and to start small [1]. Anthropic recommends finding the simplest solution possible and only adding complexity when it is needed [2]. For your first agent, that means one channel, one core task, one tool.

First: see your agent in the live demo

What does an agent sound like with your content? The EchoCall live demo shows you. You enter your website address. The demo reads your publicly reachable pages, and your robots.txt applies. After about 20 seconds an agent is ready that answers from exactly that content. You can call it in the browser or write to it, with no signup and no credit card.

The demo agent takes one call of up to two minutes, answers up to 20 messages and is deleted after 30 days. Create an account before then, take over both the chat and the voice agent along with their knowledge base and carry on with step 1 instead of starting from scratch.

If your website is missing opening hours, prices or the answer to the most common customer question, you will notice straight away.

Step 1: Build the knowledge base

Everything your agent says rests on the knowledge base. The method behind it is called retrieval-augmented generation, or RAG. It connects the language model to a searchable collection of documents [3]. For each question, the agent finds matching passages in your material and phrases the answer from them.

An EchoCall agent's knowledge base can take:

• Files in PDF, TXT, Markdown or CSV format, and JSON for voice agents. Typical examples are price lists, service descriptions, directions and forms.
• Single web pages by URL, re-read at an interval you choose.
• Your whole website via a crawl, also with a refresh interval.
• Free text for rules that are not published anywhere: "Always pass emergencies to the on-call service", "Appointments Monday to Thursday only".

Each agent has its own knowledge base. If you use voice and chat, you maintain both.

OpenAI advises using existing material for an agent's instructions: operating procedures, support scripts or policy documents [1]. Also ask your team for the 20 questions that come up every week.

What does not belong: old campaign pages and outdated prices. An agent that promises an expired discount is worse than none.

A good knowledge base lowers the risk of wrong answers but does not remove it. The taskforce of the European data protection authorities writes in its report on ChatGPT that, due to the probabilistic nature of the system, it can also produce made-up outputs [4]. So in step 2, spell out what the agent should do when it does not know something.

Step 2: Set instructions, tone and limits

Now the agent gets a role and clear rules, in an instruction text written in plain language. A structure that works well has five paragraphs:

1. Who you are: name, company, task. "You are Lena, the digital assistant of Dr Berger's dental practice in Cologne."
2. How you speak: formal or informal, tone, length of answers. On the phone: short sentences, no long lists read aloud.
3. What you do: three to five core tasks, such as booking appointments, giving opening hours, taking callback requests.
4. What you do not do: no medical assessments, no price promises outside the list, no information about other customers.
5. When you hand over: for emergencies, for complaints and for any question you cannot answer from the knowledge base.

Two points always belong in it in 2026. First, the AI notice: Article 50 of the AI Act has applied since 2 August 2026. People who interact directly with an AI system must be told [5]. In the EchoCall chat widget, the notice appears automatically in German, English or French. On the phone, you write the sentence into the agent's first message yourself. The editor shows a note about it, and the help centre has example sentences. More on this in the post on the EU AI Act.

Second, language. Decide whether the agent only speaks one language or switches to the caller's language. EchoCall supports more than 30 languages. Voice agents have automatic language detection, which you can switch off.

For the voice, listen to three or four candidates with a real sentence from your day, such as "Dr Berger's practice, good morning", not with a sample text.

Step 3: Connect tools so the agent can act

Every connected tool is an action the agent may carry out itself during a conversation. EchoCall offers, among others:

• Calendars: Google Calendar, Outlook, Calendly and Cal.com, to check free slots and book appointments.
• CRM and tickets: HubSpot, Salesforce, Zendesk, Freshdesk, ServiceNow and Jira, for example to create a contact or open a ticket.
• Data and documents: Airtable, Google Sheets, Google Drive, Notion and Confluence, to look something up.
• Notifications: Gmail, SMTP, Slack, Microsoft Teams or Discord, so that your team gets a message.
• Shop: Shopify, Shopware 6, Amazon and eBay, for example for questions about an order.
• Everything else: Zapier, Make or your own webhooks. Developers use the REST API, described in the EchoHub API Guide.

Voice agents also have built-in system tools: end the call, detect voicemail, send keypad tones, detect the language and transfer to a number.

Start with one tool, usually the calendar or the callback ticket. Anthropic writes that tool definitions deserve just as much attention as the overall instructions [2]. So state clearly when the agent should use the tool and what details it must ask for first.

Step 4: Choose the channel and set up handover to people

Voice and chat are two products at EchoCall, each with its own packages. You can start with one and add the other later.

For a voice agent on the phone, you need a phone number. Numbers and telephony costs are not included in the package prices. Many businesses keep their familiar number and set up call forwarding, for instance only when nobody answers. On a German mobile phone, the delay before forwarding can be set in five-second steps from 5 to 30 seconds [6]. Whether forwarding costs extra on your tariff is in your contract. The advantage of forwarding on no answer: nothing changes in day-to-day work, and the agent only takes the calls that would otherwise be lost.

For a chat agent, you add a script snippet to your website or install the WordPress plugin. A domain whitelist sets which sites the widget may run on. Since October 2026, visitors can also talk to the chat agent by microphone in the widget. A notice card appears first, and the minutes come from your voice allowance.

Then the handover. OpenAI names two typical triggers for a person to take over: when the agent fails at a task after several attempts, and for sensitive or irreversible actions such as cancelling orders or making payments [1]. At EchoCall, your team takes over chat conversations in the Live Inbox. The voice agent transfers the call to a number you specify. Check that someone is actually reachable on that number during the hours the agent works.

Step 5: Test, go live, refine

Anthropic recommends extensive testing in sandboxed environments, along with the appropriate guardrails [2]. A test list that has proven useful:

• Ten typical requests, phrased the way customers really say them. Not "appointment booking" but "Hi, yes, I'd need an appointment, is there anything next week in the afternoon?"
• Three awkward cases: a question that is not in the knowledge base, an angry caller and someone who changes the subject mid-sentence.
• One handover case: ask for a person and check that the transfer or the Live Inbox works as planned.
• One test of the AI notice: does it come at the start, and is it clear?

Then you go live, and the most important phase begins: the first week. Read a sample of conversations every day. Questions nobody thought of and long-winded answers are fixed in the knowledge base or the instructions. OpenAI sums it up: start small, validate with real users and grow capabilities over time [1].

In the first weeks, count three things: how many requests were completed without handover, how many appointments or tickets the agent created itself and which questions it could not answer. How high your rate without handover will be cannot be predicted reliably in advance. It depends on your requests and your knowledge base.

The five most common mistakes with a first AI agent

1. Too much at once: ten tasks, five tools and two channels on day one.
2. No way out: an agent without handover keeps customers going round in circles.
3. Outdated knowledge: use the intervals for URLs and the website crawl.
4. No AI notice: since August 2026, that breaches Article 50 of the AI Act [5].
5. Data protection after launch: an agent processes names, phone numbers and requests. Clarify the data processing agreement and the hosting location beforehand and, separately, where the language models process the conversations. EchoCall runs the EchoHub in IONOS data centres in Frankfurt and Karlsruhe. What to check with any provider is in the GDPR guide for voice agents.

No-code platform, low-code builder or custom build?

Searching for how to build an AI agent leads to three very different routes. The table is our assessment, not a measurement.

For the phone and website chat, a platform is usually the shortest route. A comparison of providers is in AI Phone Agents Compared 2026.

How much does an AI agent you build yourself cost?

EchoCall charges no setup fee. The packages according to the pricing page [7]:

For the voice agent, that is €0.30 per minute, and in the smallest chat package €0.40 per conversation. Phone numbers and telephony costs come on top. To try it out there is a trial: 30 voice minutes and 30 chat conversations for 14 days, and the credit card is verified with €1.

A worked example with stated assumptions: your agent takes 100 calls a month, each lasting two and a half minutes on average. That is 250 minutes and fits the €75 package, so 75 cents per call before telephony costs. Whether it pays off depends on how many of these calls the agent completes and what a missed call costs you. Work through your own scenario in the ROI calculator or read the ROI analysis for AI voice agents.

Frequently asked questions

Do I need programming skills to build an AI agent?

No. On a no-code platform you set up the knowledge base, instructions, voice and tools in the browser. You only need programming skills if you want to connect your own systems via the REST API or your own webhooks.

How long does it take to build an AI agent?

You can see a first agent built from your website in the live demo after about 20 seconds. For an agent with a tidy knowledge base, clear rules and one tool, plan a few hours. Fine-tuning based on real conversations then runs over the first weeks.

Can I build an AI agent with ChatGPT?

ChatGPT is useful as a chat window for staff, but an agent for your customers needs more: a maintained knowledge base, tools, handover to people, the AI notice and contracts. The differences are covered in ChatGPT in Customer Service 2026.

What happens to questions the agent cannot answer?

You define that in the instructions. It makes sense for the agent to state the gap openly and hand over: in chat to your team in the Live Inbox, on the phone by transferring the call. Wrong answers cannot be ruled out completely. So read conversations regularly and close gaps in the knowledge base.

Can I keep the agent from the live demo?

Yes, if you create an account before the demo agent is deleted after 30 days. You then take over both the chat and the voice agent along with their knowledge base and continue there.

Sources

1. OpenAI, A practical guide to building agents: cdn.openai.com
2. Anthropic, Building effective agents (19 December 2024): anthropic.com
3. Lewis et al., Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks (2020): arxiv.org
4. European Data Protection Board (EDPB), Report of the work undertaken by the ChatGPT Taskforce (May 2024): edpb.europa.eu
5. Regulation (EU) 2024/1689 (AI Act), Article 50: eur-lex.europa.eu
6. teltarif.de, GSM control codes for call forwarding (in German): teltarif.de
7. EchoCall pricing: echocall.de/en/pricing

All links were checked on 6 October 2026. This post is not legal advice.

Read next

• What Is an AI Agent? Definition, How It Works and Examples
• AI Chat Agents vs. Traditional Chatbots: Differences, Costs and When a Rule-Based Bot Is Enough
• ChatGPT in Customer Service 2026
• EU AI Act 2026: What the AI Disclosure Duty Means for Voice and Chat Systems
• AI Chatbot on WordPress in 5 Minutes