Setting up agents in Spark is more than creating a chatbot. You’ll define where the agent is active, how it responds, what instructions it follows, what actions it can trigger, and what compliance messaging it must display.
Below is a practical, step-by-step workflow based on the Spark agent setup flow.
Add and configure an agent in Spark
- Open the Spark dashboard.
- Go to Spark chat.
- Click “Add agent” to create one or multiple agents.
After the agent is created, you’ll configure two core things: which channels it’s active on and how it decides when to respond.
Select the channels the agent should be active on
In the agent setup, choose which channels the agent should work in. Use the option to add channel (the transcript describes it as “Add channel”) so you can enable the agent on specific places within Spark.
A key requirement called out in the walkthrough is that you must tell Spark which channels the agent should be active on.
Choose a response mode
Spark lets you control whether the agent responds immediately or only when there’s no existing answer.
You can choose between:
- Respond to every message
- “Unanswered question” mode (answer only when no answer exists)
If you select “Unanswered question” mode, you can also set:
- Delay minutes: how many minutes later the bot should try again
- Optional moderator-related settings (the walkthrough mentions moderator settings as something you can optionally adjust in this mode)
Customize prompts with rules (and human handoff behavior)
Spark strongly recommends adding a custom prompt.
In the transcript, the custom prompt is described as something that Spark adds at the beginning of the model, so it acts as early, top-priority instructions for how the agent should behave.
Use the custom prompt to include your operational rules, such as:
- How the agent should behave in its role
- When the agent should ask for human help (the prompt is where you encode these instructions)
The overall goal is to make the agent’s behavior consistent with how your team wants Spark to respond in real conversations.
Set conversation style and response length
Once channels and response mode are set, you can tune the agent’s communication.
Conversation style
For Gemini-based agents, Spark includes a conversation style option. The transcript lists styles you can specify, including examples like funny, kind, friendly, or casual.
Response length setting
Spark also allows you to choose how long the agent’s responses should be.
Common options described include:
- Three sentences
- Five sentences
- Let the AI decide
The walkthrough specifically notes that short requests can be handled more concisely so the responses match typical human communication.
Create AI rules and action points
To go beyond simple question-answering, Spark supports AI rules that trigger action points based on what the agent detects or determines.
The transcript emphasizes that you can define rules and then connect them to action points.
Action point example: spam detection
One described scenario is:
- If the bot thinks a message is from a spammer
- Then the action point is to gently alert everybody that the message is spam
Action point example: booking prompts
Another described scenario is:
- When users ask a relevant question
- The action point can direct users to book
These rules help the agent handle operational tasks and improve routing of user intent—without requiring every case to be manually monitored.
Configure safety and compliance disclaimers
Spark agent setup also includes safety and compliance elements.
Add disclaimers
You can add safety/compliance disclaimers such as a message indicating the bot nature (the transcript includes wording like “I’m a bot…” as an example).
The walkthrough also notes that you can tailor disclaimers for different environments, including:
- Financial institutions
- Gambling servers
Display disclaimers only at the start
A specific configuration detail from the transcript:
- Disclaimers should be configured to display only at the very beginning of a conversation with the user.
This keeps the conversation cleaner while ensuring the required message is shown upfront.
Use built-in guard rails
The walkthrough notes that Spark provides built-in guard rails by default.
The described purpose of these guard rails includes preventing users from tricking the bot into behavior such as:
- Tagging everyone
- Repeating unauthorized URLs
This helps keep the agent aligned with safety expectations and reduces common abuse patterns.
Set up a primary agent for auto onboarding
In addition to your main agent behavior, Spark supports an onboarding pattern using a primary agent.
Enable onboarding via a private channel
Set one agent as the primary agent to enable auto onboarding.
The transcript describes the onboarding as:
- Spark opens a private channel
- Spark greets this user
Auto-delete private channels after inactivity
The primary agent onboarding settings can also automatically delete private channels after a period of inactivity.
The walkthrough specifically mentions:
- Auto-delete after 24 hours of inactivity
Where settings are saved
Finally, the transcript notes that these onboarding settings are saved in the agent section.
Conclusion
To set up Spark agents effectively, start with the fundamentals—channels and response mode—then strengthen quality with a custom prompt. Next, use conversation style and response length to match how users expect the agent to communicate, and add AI rules with action points for operational behavior like spam alerts or booking guidance. Finally, configure safety and compliance disclaimers, rely on built-in guard rails, and optionally set a primary agent for private-channel onboarding with an initial greeting and auto-delete after 24 hours of inactivity.
