How to build an AI agent without code: the method and the best tools
What an agent is, eight steps from one task to a live agent, and six no-code builders priced with the unit each one counts.
Short answer: pick one repetitive task, write down how you do it by hand, then rebuild it in a no-code agent builder as a trigger, a set of instructions, a model and two or three connected apps. Run it on ten real cases with a human approving every output before you let it act alone. A first agent takes an afternoon to a week.
For the tool: Zapier Agents is the easiest start (400 agent activities a month free); Gumloop ($37 a month, 20,000 credits) suits research and data agents; n8n (from $20 a month billed annually) gives the most control for the price; Lindy works as an assistant for your inbox and meetings. If you want the full market, see our ranking of AI agent platforms or the best AI agents for small businesses.
The ranking is editorial: a sponsored placement is always labeled on its card, and no affiliate link changes a score.
6 no-code builders to create an AI agent, side by side
An AI agent is a program that receives a goal, decides the steps itself and uses tools to carry them out. A classic automation follows fixed steps you drew in advance ("when a form arrives, add a row"); an agent reads the situation and chooses ("read this email, decide if it is a sales lead, look the company up, then draft a reply or alert a human"). For the vocabulary, see our AI agent definition.
Every agent has the same parts: a trigger (an email, a form, a schedule, a chat message), instructions (the prompt that describes the job and its rules), a model (the language model that reasons, such as GPT, Claude or Gemini), tools (the apps it may read and write: Gmail, a CRM, a sheet, web search), sometimes a memory (what it remembers between runs) and guardrails (what it may never do, and when it hands over to a human). The agent loops: it picks an action, calls a tool, reads the result and decides the next step until the goal is met.
Before choosing a builder, check five things: which apps it connects to that you already use, whether it asks for approval before outside actions, which model it uses and who pays for the tokens, how it logs each run, and above all its billing unit. Each builder counts something different: Zapier counts activities (each action an agent takes; tasks once an agent moves into a Zap), n8n counts executions (one per full run), Lindy, Gumloop and Dust count credits (a currency that varies with the work and the model), and Botpress counts conversations (an exchange with at least two user messages). A ten-step agent costs one execution on n8n and roughly ten activities on Zapier.
Other names from the guides we read: Relevance AI now shows only an Enterprise plan on its pricing page, with prices through sales; MindStudio, now renamed Remy, costs $99 a month after a 30-day trial; Voiceflow shows no amounts on its pricing page (a free trial and usage-based billing for agencies, a quote for businesses), and Voiceflow alternatives covers its rivals; Make gives 1,000 free credits a month and is covered in our Make review; and Microsoft Copilot Studio sells 25,000-credit packs at $200 a month per tenant, a fit for companies already on Microsoft 365. For developers who want code instead, see agentic AI frameworks.
| Best for | Free plan | Team size | Visit | ||||
|---|---|---|---|---|---|---|---|
| 1 | Zapier Agents | Best first agent for someone who already uses Zapier | 3.8/5 | Agents Free: 400 activities a month; Agents Pro $33.33 a month billed annually ($400 a year), 1,500 activities | ✓ | Solo users and small teams on Zapier | Visit → |
| 2 | Gumloop | Best for research and data agents | 3.9/5 | Pro $37 a month with 20,000 credits; 14-day free trial; Enterprise on quote | — | Growth and ops teams | Visit → |
| 3 | n8n | Best value and control, if you accept a canvas | 4.2/5 | Starter $20 a month billed annually ($24 monthly), 2,500 executions; Pro $50 ($60), 10,000; free self-hosted Community Edition | — | Technical users, high volume | Visit → |
| 4 | Lindy | Best assistant agent for inbox, calendar and meetings | 3.5/5 | Free: $50 in credits for 7 days; Team from $29.99 per user per month for 3,000 credits | ✓ | Founders and managers | Visit → |
| 5 | Dust | Best for agents that answer from company knowledge | 3.8/5 | Free: 500 credits lifetime; Pro $24 per seat per month billed yearly ($30 monthly), 8,000 credits | ✓ | Teams of up to 100 with shared docs | Visit → |
| 6 | Botpress | Best for customer-facing chat agents | 3.6/5 | Free: 100 conversations, no top-ups; Plus $150 a month billed annually ($189 monthly), 250 conversations a month | ✓ | Businesses with customer chat | Visit → |
Is your tool missing from this page? Email us: hello@hackceleration.com
Zapier Agents
Zapier Agents lets you describe the job in plain English, pick the apps the agent may use, and let it decide the steps. Its strength is the catalog: the same app connections as Zapier's Zaps, so almost any tool you already use is reachable. The Agents Free plan runs up to 400 activities per month; Agents Pro costs $33.33 a month billed annually ($400 a year) for 1,500 activities. An activity is any billable action the agent takes: using a trigger, answering from a knowledge source, running an action, browsing a page or searching the web each count as one.
Activities are a separate meter from the tasks your Zaps use, so agent usage does not eat your Zap quota, and the Zapier Free plan's 100 tasks a month stay for your classic automations. Two limits matter on the free plan: tests count toward the monthly quota, and a single run stops at 10 activities (40 on Pro), after which the agent asks you before going on. At the monthly limit, agents stop running but you can still chat with them.
Know where the product is heading: Zapier is migrating Agents into AI by Zapier, an AI step inside the Zap editor, billed in tasks like any other Zap step. An AI step costs 1, 3 or 5 tasks depending on the model tier, a run pauses for your review once it passes 75 tasks (adjustable up to 500), and you can require approval tool by tool. Zapier has not set a date to turn off Agents and will email users before it does. Compare it with the other automation tools in n8n vs Zapier, or read our Zapier review.
- 400 activities a month on the free plan
- Activities do not consume Zap tasks
- Emails at 80% and 100% of the quota
- ✓9,000+ app integrations
- ✓Plain-English setup
- ✓A free plan large enough to build a real agent
- ✗Tests count toward the free quota
- ✗10 activities per run on the free plan
- ✗Being folded into AI by Zapier, billed in tasks
Start here if your agent mostly moves information between apps you already use.
Gumloop
Gumloop is built around agents that chain web research, data enrichment and model calls. One plan is public: Pro at $37 a month, with 20,000 credits a month (7,400 plus 12,600 bonus credits), unlimited agents, unlimited seats and teams, 5 concurrent workflow runs and 25 concurrent agent chats. A 14-day free trial lets you build before paying.
A credit is Gumloop's single currency for model tokens and paid tool calls, indexed at $0.005. On top of the list cost of each task, Gumloop adds an 8% orchestration fee; if you plug in your own model API keys, the token credits drop to zero but the fee rises to 16% of what the task would have cost. When you hit 20,000 credits mid-month, your agents stop until you enable overage, billed at $0.005 a credit. Unused credits do not roll over. If Gumloop does not fit, our page on Gumloop alternatives lists the other options.
- 20,000 credits a month on Pro
- Unlimited seats on one plan
- 35+ models available
- ✓One plan for a whole team
- ✓Strong at research and enrichment
- ✓Overage you switch on yourself
- ✗A 14-day trial, then $37 a month
- ✗8% fee on top of usage, 16% with your own keys
- ✗Agents stop at the credit limit without overage
The best fit for agents that gather and process data, as long as you watch credits per run.
n8n
n8n is a workflow canvas with an AI Agent node: you plug a chat model, a memory and tools into it, and the agent decides which tool to call. When a node is not enough, you can drop in a JavaScript or Python code step. It is also the least "no-code" tool here: expressions and JSON appear quickly. Cloud plans are priced by executions: Starter costs $20 a month billed annually ($24 month to month) for 2,500 workflow executions, Pro $50 ($60) for 10,000. Business, at $800 ($960) for 40,000, is sold for self-hosting only. Prices are on our n8n pricing page.
An execution is a single run of the entire workflow, however many steps it has, so a long agent costs the same execution as a short one: the cheapest unit on this page for multi-step work. The model is not in the plan price: you pay it on your own API key, or, on Starter and Pro, from a prepaid balance of Gateway credits that opens OpenAI, Anthropic, Google and other models without keys. Starter stops any single execution after 5 minutes (40 on Pro), and each tool call can require a human approval. A self-hosted version is on GitHub under the Sustainable Use License, which allows internal business or personal use, not reselling n8n as a service. For a ready-made example, see an n8n agent that schedules meetings from Gmail and Google Calendar. See our n8n review and n8n alternatives.
- One execution per run, whatever the length
- Choose any model and memory
- Human approval for tool calls
- ✓One execution per run, however many steps
- ✓Full control of tools and memory
- ✓Large template library
- ✗Steeper learning curve than the others
- ✗Model costs billed separately
- ✗License is not open source
Our pick when the agent will run hundreds of times a month and you can spend a weekend learning the canvas.
Lindy
Lindy is an assistant you brief in plain language and that works in your inbox, calendar, meetings and Slack, with 3,000+ integrations. You start with $50 in free credits that last 7 days; after that, the Team plan starts at $29.99 per user per month for 3,000 credits, and top-ups cost $10 per 1,000 credits.
Credits measure the work done: an everyday ask (a summary, a lookup, a drafted reply) costs 2 to 250 credits, deep work such as a competitor report 250 to 1,000, and a big build 1,000 to 2,500. So 3,000 credits cover at least 12 everyday asks, or between 3 and 12 competitor reports. Every seat adds its credits to one pool the whole team draws from. Monthly credits do not roll over; when the pool runs out, Lindy pauses instead of billing more. Anything with outside impact, such as sending an email, waits for your approval. Read our Lindy review, the detail of Lindy pricing or compare it in Lindy vs MindStudio.
- $50 of free credits for a week
- Approvals before outside actions
- 3,000+ integrations
- ✓Setup in about 2 minutes from a plain-language brief
- ✓Pauses instead of surprise billing
- ✓Pick your model per task
- ✗Every active user is billed, Slack mentions included
- ✗Credit cost per task varies widely
- ✗Monthly credits do not roll over
Pick it for personal and team assistant work; for data pipelines, Gumloop or n8n fit better, and Lindy alternatives lists the rest.
Dust
Dust connects your company's knowledge (documents, wikis, conversations) and lets anyone build an agent on top of it, with a choice of 20+ models from OpenAI, Anthropic, Google, Mistral and DeepSeek. The Business plan is built for teams of up to 100 people. The free plan gives 500 credits for the lifetime of the account, enough to try a few agents. Pro costs $24 per seat per month billed yearly, or $30 month to month, with 8,000 credits per seat per month; Max costs $120 billed yearly ($150 monthly) for 40,000 credits.
Each seat's monthly credits reset at the start of the billing period and do not roll over. Calling a Dust agent from outside, through its API or from automation platforms such as Zapier, Make, n8n or Power Automate, is billed at $0.01 a credit. Adding a member mid-cycle is prorated, and the new member gets the full credit allowance at once. When a Pro or Max seat runs out, it can keep going on workspace overage only if an admin has turned it on, up to a cap. For pure outbound automation, Zapier or n8n fit better.
- 8,000 credits per seat on Pro
- Agents over your own documents
- Callable from Zapier, Make, n8n
- ✓Good for internal knowledge agents
- ✓Simple per-seat pricing
- ✓Model choice
- ✗Free credits are lifetime, not monthly
- ✗Cost grows with headcount
- ✗Programmatic calls billed per credit
A strong choice for an internal assistant that answers from your company docs.
Botpress
Botpress builds agents that talk to your customers on a website chat, WhatsApp and other channels. The Free plan includes 100 conversations, 3 AI agents and 3 seats, with no top-ups or overages. Plus costs $150 a month billed annually ($189 monthly) for 250 conversations and $25 of AI usage included; Team costs $750 billed annually ($939 monthly) for 1,500 conversations.
A conversation is any exchange with at least two messages from the end user in the billing month, and a conversation that spans two months counts in both. On paid plans, once you reach 95% of your quota, a pack of 100 conversations is added automatically: $65 on Plus, $50 on Team. This auto-recharge cannot be turned off, and a pack stays on the bill month after month until you remove it in billing settings. Model spend is a second meter: at 95% of the AI spend limit, Botpress buys a $10 credit grant for you, sold at cost. Our page on Botpress alternatives covers the other chat builders, and AI agents for customer support ranks the field.
- 100 conversations on the free plan
- Website, WhatsApp and other channels
- 3 seats on the free plan
- ✓Built for customer conversations
- ✓Predictable conversation-based price
- ✓Real free plan to launch a small bot
- ✗Paid plans start high, at $150 a month billed annually
- ✗Automatic packs on paid plans
- ✗Steeper learning curve
The right tool when your agent's job is talking to customers, not moving data.
How to build an AI agent in 8 steps
The order we recommend for a first agent, from the task to a month of monitoring. Plan an afternoon for a simple agent and up to a week of evenings for one that touches customers.
The steps
1. Pick one task that repeats every week
Choose a task you do at least weekly, that takes 20 to 60 minutes and follows a pattern: sorting support emails, qualifying inbound leads, writing a meeting follow-up, compiling a weekly competitor brief. Write the job in one sentence with the trigger, the decision and the action: "When a form arrives, score the lead against our ideal customer and alert sales on Slack if it scores above 7." A rare task is a poor first agent: you will not see enough runs to fix it.
2. Write down how you do it by hand
List every step you take today, in order, with the information you look at and the rule you apply. These notes become the agent's instructions. Mark the points where you use judgment: they are where the agent will need the clearest rules or a human check. If you cannot describe the rule, the agent cannot follow it.
3. Check that you need an agent, not an automation
If every case follows the same steps, a classic automation (a Zap, an n8n workflow, a Make scenario) is the better fit. You need an agent when the input varies and a decision is involved: reading free text, choosing between several actions, searching for missing information. Many good setups mix both: fixed steps, with one agent step where judgment is needed.
4. Choose the builder by where the work happens
Pick from the apps you already use and the billing unit that fits your volume. Moving data between business apps: Zapier Agents. Research and enrichment: Gumloop. High volume and full control: n8n. Inbox, meetings and Slack: Lindy. Answers from company documents: Dust. Customer chat on a website or WhatsApp: Botpress. Start on the free plan or trial to build, not to launch.
5. Write the instructions and connect two or three tools
Give the agent a role, the steps, the rules and the output format: "Classify each email as Urgent, Routine or Spam. For Routine, draft a reply under 150 words in our tone. For Urgent, post the email to #support with a one-line summary. If you are unsure, choose Urgent." Connect only the apps the job needs: each extra tool adds a way to fail and, on n8n, more tokens per turn. Give read-only access where you can, and add memory only if the agent must remember past runs, such as a lead follow-up over several weeks.
6. Run it on ten real cases with a human approving
Use real history, not invented examples: last week's emails, last month's leads. Keep the agent on draft or approval mode and compare each output with what you would have done. Write down every failure: a wrong category, a tone that is off, a field left empty. Expect a first version to be wrong on a fair share of cases; most of the gain comes from two or three rounds of prompt fixes.
7. Add guardrails and a way out
Set what the agent may never do (send money, delete records, email a whole list), a cap on steps per run, and an explicit "not sure" path that routes to a person. Use the approval features the builders offer: Lindy waits for approval before any action with outside impact, and Zapier's free plan caps a run at 10 activities. n8n and AI by Zapier let you require approval tool by tool. On n8n, also set the agent's maximum iterations so a failing tool cannot loop.
8. Launch, watch the logs and the bill for a month
Turn it on for a small share of the work first, then widen. Read the run logs every week for the first month and fix the recurring failures in the prompt. Watch usage too: note how many credits, activities or executions a typical run uses after the first week, multiply by your monthly volume and check it against your plan before the end of the cycle. Remove the human check only once the agent has run cleanly for a few weeks.
8 pricing traps the tutorials skip
From the builders' pricing pages, help centers and licenses, and from users' reports.
The traps
Do not build on Zapier's free quota
On the Zapier Agents free plan, tests count toward the 400 monthly activities, and a single run stops at 10 activities. Twenty test runs of a ten-step agent use 200 activities, half the month, before the agent does any real work. On paid plans, testing does not count. Build with short test cases, or upgrade for the build week. If you build the same agent as an AI by Zapier step instead, full-Zap tests currently consume tasks like production runs.
Your own API keys can cost more on Gumloop
Gumloop adds an 8% orchestration fee to the list cost of each task. Bringing your own model keys removes the token credits but doubles the fee to 16%, still calculated on the full list cost. On a task with a list cost of 1,000 credits, all of it model tokens, you pay 1,080 credits with Gumloop's models, or 160 credits of fee plus your own provider's bill with your keys: worth it only if your provider charges you clearly less than Gumloop's list price.
Cap your agent's loops before the first run
An agent that calls a failing tool can retry forever. A 2024 user report describes an n8n agent whose tool returned a 404 error and looped, for a bill of $300 in 30 minutes on the model provider. Set a maximum number of iterations, and a hard monthly spending cap at your model provider: the builder's plan does not cover tokens you pay directly.
Each attached tool adds tokens to each turn
Each tool you attach to an agent sends its description to the model at every step. A 2026 measurement on n8n found that four tools on one agent added about 3,200 input tokens per turn. Detach the tools a given agent does not need, and split a do-everything agent into two smaller ones.
n8n's self-hosted version is not open source
The self-hosted n8n on GitHub is under the Sustainable Use License: you may use it for your own internal business purposes or for non-commercial use. If you plan to sell agents built on n8n to clients as a hosted service, read the license first or budget for a commercial agreement.
n8n Starter stops a run after 5 minutes
On n8n Cloud, Starter caps a single execution at 5 minutes, Pro at 40 minutes, and the free trial at 180 seconds. A long research agent that browses pages and waits on a slow model can hit the Starter limit and fail halfway. Time your longest run during the trial; if it gets close to the cap, budget for Pro at $50 a month billed annually rather than splitting the agent in two.
Botpress counts a conversation twice across months
A Botpress conversation that spans two billing months counts in both. On paid plans, a pack of 100 conversations is added automatically at 95% of your quota ($65 on Plus), and model spend triggers its own $10 grants at 95% of the AI spend limit. Test chats in the emulator count as conversations too, and the auto-added packs cannot be switched off. Budget two meters, not one.
Credits do not roll over, and seats add up
Lindy, Gumloop and Dust all reset monthly credits at the start of the billing cycle: what you did not use is lost. Lindy also bills every active user, including a teammate who only mentions it in Slack once their 7-day trial ends, and Dust prices per seat. Count the people who will touch the agent, not only the one who builds it.
Building an AI agent: your questions
Can I build an AI agent for free?
Yes, for a small volume. Zapier Agents runs 400 activities a month on its free plan, Botpress 100 customer conversations, and Make gives 1,000 credits a month on its free plan. What is rarely free is the model: n8n and Make agents that call OpenAI or Anthropic on your own key pay the provider per token, and a ChatGPT Plus subscription does not include API access.Do I need to know how to code to build an AI agent?
No, but you need to think in steps and rules. Lindy and Zapier take a plain-English brief; n8n shows expressions and JSON early, and lets you add JavaScript or Python code steps when a no-code node is not enough. Most beginners get stuck on the logic of the task, not on code: write the manual steps first (step 2 of the method).How long does it take to build an AI agent without coding?
An afternoon for a simple agent, up to a week of evenings for one that touches customers. Most of the time goes into testing: in a published case, a support agent built on 200 help articles answered 73% of test questions correctly on the first try and 89% after about two hours of prompt and knowledge-base fixes.Can ChatGPT build an AI agent for me?
It can help you write one, but it is no longer a builder for most people: personal ChatGPT accounts (Free, Go, Plus and Pro) cannot create new custom GPTs, and OpenAI plans to retire custom GPTs in favor of plugins. Use ChatGPT to turn your manual steps into a clear prompt and a list of rules, then paste them into a builder that has triggers and app connections.What is the difference between an AI agent and a chatbot?
A chatbot answers in a conversation; an agent takes actions to reach a goal. A support chatbot replies from a knowledge base; a support agent can also open the ticket, look up the order in your store and issue the return label. Botpress and Dust sit closer to the chatbot side, Zapier Agents and n8n closer to the action side.Is it safe to let an AI agent send emails for me?
Only after a supervised period. Keep it on draft or approval mode for the first weeks, and give it write access to as few apps as possible. Lindy asks for approval before any action with outside impact by default; n8n and AI by Zapier let you require approval for each tool, so you can leave reads free and gate the send action.Which AI model should my agent use?
Start with the builder's default, then switch only with a reason. On n8n you choose the model and pay for it outside the plan, on your own key or from prepaid Gateway credits; on Lindy and Gumloop the model's cost is turned into credits, so a larger model uses more credits for the same task.Can I sell AI agents to clients without coding?
Yes, but check the terms of the builder first. n8n's self-hosted license allows internal business use only, so a hosted service for clients needs another arrangement, and Lindy bills every active user in a workspace. Voiceflow sells an offer built for agencies, with multi-client workspaces, white-labeling and client handoff tools. For client work, see AI agents for agencies.
