You can design and publish a Coze chatbot without first writing a conventional application: create a visual workflow, add and connect nodes, configure inputs and outputs, test it, publish it, and attach it to an agent. Coze documents a Free plan with 10 credits per day, but that does not mean unlimited free GPT-4-family usage. The number of calls depends on the model and deployment channel, so check the current credit table and your account options before launching.
What you can build in Coze
Coze combines an agent with configurable model and prompt settings, plugins, visual workflows, knowledge, memory, and chat-experience options. A workflow is the visual logic that carries out a task; the agent provides the conversational interface and can call that workflow. Coze describes knowledge as retrieval-augmented generation over uploaded text, table, and image content. Coze’s workflow guide defines a workflow as “a set of executable instructions used to implement business logic or complete specific tasks.”
This tutorial uses the documented workflow-to-agent route. It is suitable for a small task such as routing a question, formatting an answer, or combining a prompt with a plugin. The exact nodes depend on what you want the bot to do; the official guide establishes the editor process, not a particular prebuilt GPT-4 template.
Does “free” mean free GPT-4 chatbot usage?
No. Coze’s subscription documentation lists the Free plan at 10 credits per day. In its model-and-channel estimate table, GPT-4o mini is shown at 100 calls per day for the Coze Store/Telegram/Discord/Instagram/Messenger/Slack/LINE/WhatsApp grouping. Other GPT-4o and GPT-4 Turbo variants have lower estimated daily call counts in some entries because models consume different amounts of credit. Those figures are Coze’s estimates, not a universal allowance for every model or channel. The subscription page does not state a publication year; figures here reflect the page as checked on October 4, 2026. Check Coze’s current plan and channel table before relying on an estimate.
#1 Best Overall
Free workflow creation and ongoing inference are different questions: a builder may prototype without a paid subscription and still encounter daily credit or interaction limits. Channel eligibility and billing also differ. Coze says paid-plan credits apply only to specified channels, while some other channels require separately purchased credit messages; interaction limits may apply even where credits are not consumed. Do not assume a bot is free to operate simply because you could build it without paying.
Coze’s documentation lists Premium Lite at 100 credits per day, Premium at 400, and Premium Plus at 1,000. These are daily plan credit figures from the same subscription page, not guarantees of a particular number of conversations: the credit cost varies by model, and channel-specific terms apply. Choose a plan only after checking the live table for your intended model and deployment channel.
Rank #2
Create a workflow in Coze
- Open the workflow editor: Sign in, choose a workspace, open Library, create a Workflow resource, and give it a name. The editor opens with start and end nodes. Coze’s documented workflow setup uses this sequence.
- Add the task logic: Add nodes to the canvas and connect them in the order the task should run. Coze’s guide describes built-in basic nodes and plugin nodes. Use only the nodes needed for the result you want; the workflow should pass the appropriate data from one step to the next.
- Configure node data: Set each node’s inputs and outputs so that the next node receives the expected values. Inspect the start and end of the flow as well as intermediate nodes: the workflow needs a clear input and a useful returned result.
- Run a test: Test-run the workflow with representative input and inspect node results. If a node has missing or unexpected data, correct its input/output configuration or connections and run the test again.
- Publish the workflow: Publish only after the test output is usable. Coze requires the workflow to be published before it can be added for use within an agent.
Attach the workflow to an agent
- Open or create the agent in the same Coze workspace.
- Add the workflow: In the agent’s Workflows section, add the published workflow.
- Explain when to use it: In Persona & Prompt, refer to the workflow by its name and describe the circumstances in which the agent should call it. This helps the agent select the workflow for the intended task rather than treating it as an unexplained resource.
- Test the full conversation: Ask the agent a question that should invoke the workflow, then check that the result returned in chat is relevant and readable. Testing only the workflow does not establish that the agent will invoke it correctly.
Coze’s feature overview covers model selection and parameter configuration, prompt engineering, plugins, workflows, knowledge, memory, and chat-experience configuration. These are separate controls: a workflow defines task steps, while model and prompt choices affect how the agent handles conversation. See Coze’s feature overview for the platform capabilities described in its documentation.
Publish the chatbot to a channel or API
Coze’s feature overview lists deployment options including Discord, Telegram, Facebook Messenger, LINE, Slack, Instagram, WhatsApp, Lark/Lark Base, and Cici, as well as API and Web SDK integration. Channel availability can change, so confirm the current publishing options in your account. Match the channel to both the audience and the plan’s billing eligibility; a model’s displayed call estimate may be tied to a particular group of channels rather than all deployments.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsFor API conversations, Coze’s developer guide says the agent must first be created and published as an API service. It also notes that some creation settings are not exposed through the API. Therefore, do not expect every agent configuration step to be available through an API call; configure and publish the agent as required in Coze, then use the documented API service path. Read Coze’s API conversation guide for the current prerequisites.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Choose a model and deployment based on limits
There is no universally best GPT-4-family choice established by Coze’s plan table. Compare the current per-model credit cost and estimated calls for the channel you plan to use, then weigh the response capability and latency your task needs. A lightweight FAQ bot with modest complexity may not need the same model choice as a workflow that depends on more demanding reasoning. The documented credit estimates help with usage planning, but they do not establish a performance ranking for your particular chatbot.
Quick Recap
- For a prototype: Start with the Free plan’s current allowance and a model/channel combination shown in the official table. Test with real representative prompts and observe actual account usage.
- For a messaging deployment: Check the exact channel grouping and its credit or interaction terms. Do not transfer an estimate from one group to another.
- For an app integration: Use the API or Web SDK option only after confirming the agent’s publication requirements and which settings must be configured outside the API.
- For recurring public use: Estimate expected conversation volume against the live credit allowance and channel rules, then decide whether a paid plan or separately purchased credits are necessary.
Common setup problems to check
- The workflow does not appear in the agent: Confirm that the workflow was published, then add it in the agent’s Workflows section.
- The workflow runs but returns the wrong result: Inspect node connections and input/output configuration, then test the workflow independently before retesting through the agent.
- The agent does not call the workflow: Check that the workflow is attached and that Persona & Prompt refers to it by its name and explains when it should be used.
- The bot stops responding as expected: Check the account’s remaining credits, selected model, deployment channel, and any interaction limits. A Free-plan call estimate is conditional on its model and channel entry.
- An API setup cannot change a setting: Coze documents that some creation settings are not available through the API. Configure those settings through the supported creation flow before publishing the API service.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
