MMOOZZGG.com

How to build an AI agent

Build an AI agent without coding: 10 agents you can make yourself

7 min read13 sources

You can build an AI agent without coding from open tools: the Flowise builder under the Apache 2.0 license, n8n workflows and models that Ollama runs on your own computer. Below are 10 agents you can build without a programmer, each with its tool, build time and limit.

On this page · 4 sections

Build it yourself in an evening

How to read the list

Each agent has four lines: what it does, what to build it with, how long it takes and where its limit is. We give the build time as a level of difficulty, not in hours: “in an evening” means one task and one data source, “in a week” means several steps and a document base.

You can take an open model (a neural network that understands and writes text) and run it on your own computer with Ollama. Then the text of your documents goes nowhere, and you pay only for the computer and electricity.

Flowise builds agents with a mouse: it is a visual builder with the motto “Build AI Agents, Visually”. n8n builds automation workflows, chains of steps such as “an email arrives, the model reads it, the summary goes to you”, and places an AI agent inside them.

Dify builds AI applications with a knowledge base: you upload documents, and when someone asks a question, the system first searches them for matching passages. All four tools are open, but their license terms differ; see the end of the page.

10 agents with tools and build times

1. Answers from a knowledge base: the agent answers staff questions on internal rules and the price list and names the document it took the answer from. You build it in Flowise with “Document Stores”: upload a file, split it into chunks and press “Upsert” (save it to a vector database, where text is searched by meaning). Time: in a week; limit: messy documents give messy answers.

2. Email sorting: the agent reads a new email, sets its topic and urgency and sends you a short summary. In n8n, the “Gmail Trigger” node with the “Message Received” event starts the workflow, and the “AI Agent” node makes the decision. Time: in an evening; limit: the agent sorts and summarizes, but it does not reply to clients.

3. Meeting summaries: you paste a transcript of a conversation, and the agent picks out decisions, tasks and owners. Ollama is enough: the command “ollama run” with a model name opens a chat with the model right in the terminal window. Time: in an evening; limit: the agent works with text, so audio must be transcribed first.

4. Email drafts: the agent writes a reply in the company’s tone based on a few samples in its instructions. You build it in the same n8n workflow as email sorting, or in an Ollama chat. Time: in an evening; limit: a person sends the email, and the agent only prepares the text.

5. A Telegram bot that answers frequent questions. You create the bot in @BotFather with the /newbot command, BotFather gives you a token (a key through which the builder controls the bot), and in n8n the “Telegram Trigger” node with the “Message” event receives messages. Time: in an evening; limit: it answers only from your text, with no access to orders.

6. A guide to contracts and templates: a manager asks where the standard contract sets the payment terms and gets an answer with the clause number. You build it in Dify with the “Knowledge” base. Time: in a week; limit: the agent finds and summarizes, and a lawyer makes the decision.

7. Review analysis: you export reviews to a file, and the agent groups them by topic and shows what people complain about most. Ollama and an open model are enough. Time: in an evening; limit: you export the reviews from the platforms yourself, and the numbers are worth double-checking.

8. Product descriptions: the agent writes a product card from the specifications and your sample. Ollama and an instruction with an example are enough. Time: in an evening; limit: a person checks sizes, materials and prices, because the model may distort them.

9. First screening of resumes: the agent compares a resume with the job requirements and writes what the candidate lacks. You build it in Flowise or Dify. Time: in a week; limit: resumes hold personal data, and the rules for it are below.

10. Text review: the agent checks emails and posts against your style and a list of banned words and suggests edits. Ollama and an instruction are enough. Time: in an evening; limit: it does not check meaning or facts.

Step-by-step guides for these tools are in the section: agent on n8n and agent on Flowise. For agents 1, 6 and 9, the page about a knowledge base agent will help.

Where building it yourself ends

All 10 agents on the list prepare text, and the decision and the action stay with a person. Your own skills reach their limit where the agent starts to act by itself: write to the CRM, change an order in 1C (a Russian accounting and ERP system), reply to a client without review.

Model developers advise not to rush across this line. In December 2024 Anthropic wrote: start with a simple solution and add complexity only when the simple one fails; sometimes this means not building an agent at all.

The second limit is personal data. Under 152-FZ, Russia’s personal data law, data of Russian citizens is collected and stored in databases in Russia, and before a transfer abroad the operator files a notice with Roskomnadzor, the data regulator.

So give resumes, client chats and phone numbers only to an agent on your own computer or on a server in Russia. How we handle this is described on the data security page.

The third limit is an agent for one person versus an agent for a department. When twenty employees use an agent, you need access rights, an action log and a person responsible for its work.

What to do when the agent outgrows the experiment

The signs are simple: a whole department uses the agent, money depends on its answers, and people want to connect it to the CRM or 1C. This is already a developer’s level, and a home-made build is worth reviewing.

Start by counting: how many hours a week the agent saves and how much one of its mistakes costs. If you have no answer, you need a metric first, not the next agent.

Check the license. n8n is distributed under the Sustainable Use License (fair-code, not an OSI open license): you may install it on your own server and change it, but you may not resell it as a cloud service.

Dify uses Apache 2.0 with two conditions: without permission, you may not run a multi-tenant cloud service like Dify or remove the logo from the console. Flowise is distributed under Apache 2.0 without such conditions.

All three fit work inside a company. If you want to sell agents to your clients, read the terms before the first sale.

Whether it pays more to finish the agent yourself or order it is covered on a separate page.

Sources and fact-checking

Checked by the editors on 16 October 2026.

  1. GitHub, 16 October 2026 (github.com)
  2. GitHub, langgenius/dify, LICENSE, 2026 (github.com)
  3. Wikipedia, 2026 (en.wikipedia.org)
  4. Alpine Linux packages, 2026 (pkgs.alpinelinux.org)
  5. Anthropic, 19 December 2024 (anthropic.com)
  6. Brace Law Firm, 9 July 2023 (brace-lf.com)
  7. n8n documentation: AI Agent node, checked 16 October 2026 (docs.n8n.io)
  8. n8n documentation: Gmail Trigger node, checked 16 October 2026 (docs.n8n.io)
  9. n8n documentation: Telegram Trigger node, checked 16 October 2026 (docs.n8n.io)
  10. Flowise documentation: Document Stores, checked 16 October 2026 (docs.flowiseai.com)
  11. Dify documentation: Knowledge, checked 16 October 2026 (docs.dify.ai)
  12. Ollama documentation: README: Quickstart, checked 16 October 2026 (github.com)
  13. Telegram documentation: Bot Features: BotFather, checked 16 October 2026 (core.telegram.org)

Found a mistake? Tell us

Publications

Contact us