Chatbot template: start from the conversation workflow, finish with your content
This template starts from the conversation workflow, a Start, an agent and an End, and shows what you add to make a chatbot you can release: scope, sources, tests and an approval before it goes live.
Last reviewed 7 October 2026
Starting point
The status rests on the conversation workflow, which has one agent node in conversation mode with a timeout and retry setting, and the HR screener chatflow. The workflow has a fixed test message and does not keep history between runs, so it is a base, not a finished bot.
The library has a template, chatflow, MCP server or governed template that covers part of the job. You assemble the rest in Studio.
What this template does
A chatbot is easy to start and hard to finish. A model that chats is a few minutes of work. A bot that stays within a topic, knows when to stop, and has been tested on the questions people really ask takes longer. The owner is whoever the bot speaks for, and they are accountable for what it says to the public.
The template is the build order. You define what the bot is for and what it will refuse, give it the content it may use, and test it against a written set of questions. Only then does a named person approve its release to a live channel. It is the generic build; channel-specific flows, such as WhatsApp, are separate pages.
The workflow, step by step
Steps marked as approval gates pause the run until a named person approves. Nothing after a gate runs before that decision, and the decision is recorded.
- 01
Write the scope
Write down who the bot talks to, what it answers, what it refuses and when it hands over to a person. A page is enough. The scope becomes the opening of the system prompt and the first section of your test set. - 02
Choose the sources
List the content the bot may answer from: help pages, policies, product facts. Remove anything out of date or contradicted elsewhere. The agent is told to answer only from these, and to say plainly when they are silent. - 03
Configure the agent node
Open the conversation workflow, create an agent with your system prompt and model, and replace the fixed test message with the incoming one. Set the timeout and retries, and turn on conversation history, which the fixture leaves off. - 04
Add a structured path where needed
If the bot must collect facts, for example name, order number and issue, follow the chatflow shape: one field per question, a yes or no where that is enough, and a few wordings for each question. Required fields are asked until they are answered. - 05
Test before release
Run a written set of questions: common ones, off-topic ones, awkward ones and attempts to push the bot out of scope. Read every answer. Record the failures, fix the sources or the prompt, and run again. - 06
Release owner approves
The Studio approval step pauses the release for a named person, who reads the test results and approves publishing to the live channel. If a second person co-owns the bot, assign someone who did not write the prompt.Approval gate: a named person approves before the next step runs. - 07
Watch and improve
After release, the owner reads a sample of conversations each week. Failures become new test questions, new content or a tighter scope. Each change is tested before it goes out, and the record shows which version said what.
What it can do, cannot do, needs approval for, and records
Can
- Hold a conversation from your system prompt
- Answer from sources you supplied
- Collect named fields in a scripted path
- Keep a transcript of each conversation
Cannot
- Go live without the release approval
- Answer well from sources it was not given
- Take actions in other systems unless you wire them
- Replace a person for sensitive cases
Requires approval
- Releasing the bot to a live channel
- Changes to the scope or the system prompt after release
- Adding a tool the bot can act through
Records
- Scope, prompt and sources by version
- Test questions and results
- Release approver and time
- Transcripts of live conversations
What the library holds for this job
| Asset | Type | Covers | Role in this flow |
|---|---|---|---|
| Conversation | Workflow template | Part of the job | Start, one agent node in conversation mode, End. The agent id is a placeholder and the message is a fixed test question. |
| HR Screener - Dublin Warehouse Position | Chatflow template | Part of the job | A scripted, field-by-field conversation with yes or no questions and wording variants; it shows the structured shape, in a recruitment setting. |
| Generate FAQ Answer | Prompt | Used inside the flow | A starting prompt for question-and-answer text, useful for the first draft of your FAQ content. |
| Webhook | Integration | Used inside the flow | Receives messages from your site or app and starts the conversation run. |
Integrations this flow uses
- Webhook: Receives messages from your site or app.
- OpenAI: One model option for the agent; any model reachable through Connect can be used.
- Slack: An internal channel for testing and for handovers.
- Telegram: An example public channel for the bot.
The full list of tools Studio connects to is on the integrations page.
What you supply, and what this page does not cover
- You supply the scope, the sources and the test questions. The template supplies the order of work.
- The conversation fixture has a placeholder agent id and a fixed test message, and history is off. Treat it as a base.
- The chatflow in the library is for recruitment screening in one city. Use its shape, not its content.
- Voice, avatars and channel-specific rules are not covered.
Common questions
- Is the conversation workflow a finished chatbot?
- No. It is three nodes, Start, one agent and End, with a fixed test question. It proves an agent can converse inside a workflow. Scope, sources, memory, handover and release are what you add, in the order above.
- What is a chatflow, and when do I use one?
- A chatflow is a scripted conversation made of fields, each with a question and a type, such as yes or no. Use it when the bot must collect specific facts. Use a free conversation agent when the questions are open.
- How do I stop the bot answering things it should not?
- State the scope in the prompt, limit it to your sources, and test the edges: off-topic questions, requests for advice, attempts to override instructions. No wording is absolute, so keep a handover route and read live transcripts.
- How is this different from the WhatsApp template?
- This page is the general build, independent of channel. The WhatsApp template adds the channel: sender matching, handover triggers, commitments and opt-outs. Build the bot here first, then connect it to a channel.