Spark Chat Agents Setup Checklist: Channels, Prompts, Rules & Safety

Setting up Spark Chat agents helps you automate responses while keeping control over where the agent operates, how it answers, and how safety is handled.

Below is a practical checklist based on the Spark agents setup flow: create agents, configure channels and response behavior, shape conversation style and length, add rule-based actions, and finish with safety/compliance settings and primary agent onboarding.

Add and configure agents in Spark Chat

1. Open the Spark dashboard and go to Spark chat.
2. Click Add agent to create a new agent.
3. You can create multiple agents for different purposes (for example, separating testing vs. production-ready behavior).
4. The only required configuration is selecting which channels the agent should be active on.

That channel selection determines where users will interact with that agent, so it’s the first decision to lock in before you tune behavior.

Choose response mode, channels, and a custom prompt

After channel setup, configure how and when the agent responds.

Response mode

Spark lets you select one of two response modes:
- Reply to every message
- Respond only to unanswered questions

If you want the agent to avoid interrupting an already-answered thread, the “unanswered questions” approach is the direct fit.

Custom prompt (recommended)

Spark also encourages adding a custom prompt so the agent’s behavior matches your expectations.

In practice, this is where you define how you want the agent to interpret requests and respond within your intended boundaries.

Optional timing and moderator-related settings

The setup includes optional controls such as:
- Timing for delayed answering (so the agent can wait before responding)
- Moderator-related settings (available as part of the configuration options)

Configure conversation style, response length, and action rules

Once the agent can respond and you’ve shaped the prompt, you can refine the “voice” and the operational logic.

Conversation style (Gemini-based)

Spark supports a configurable conversation style for Gemini-based agents. You can choose a style such as:
- funny
- kind
- friendly
- casual

This setting helps standardize tone across replies, based on the type of interaction you want.

Response length setting

You can control how long responses should be:
- Set a fixed number of sentences (for example, three or five)
- Or allow the AI to decide the response length based on the request

The setup description notes that the default behavior targets concise answers for shorter questions.

Define AI action rules

Spark includes rule-based “action points” that trigger based on what the system detects.

One described example:
- If the system determines a message looks like spam, the action can be to gently alert everyone that it’s a spam message.

You can also use rules to guide responses, such as:
- Directing users to a resource when they ask something

These rules are designed to turn detection into consistent, controlled next steps rather than letting the agent improvise.

Implement safety and compliance disclaimers and guard rails

Safety is handled through a combination of disclaimers and built-in safeguards.

Add a bot disclaimer

Spark lets you configure a disclaimer such as a statement that the user is interacting with a bot.

The setup also references additional disclaimer options related to certain environments (including financial institutions and gambling servers).

Control where the disclaimer appears

You can choose whether the disclaimer appears:
- at the beginning of a conversation

This helps ensure users are aware of bot involvement without repeating the message throughout every exchange.

Use built-in guard rails

The configuration includes safeguards intended to reduce common abuse scenarios, such as:
- preventing users from tricking the bot into tagging everyone
- preventing the bot from repeating unauthorized URLs

The guard rails are described as being enabled by default, with safeguards provided for you.

Enable a primary agent with auto-onboarding

Finally, Spark supports a primary agent setup that can be enabled for a guided onboarding experience.

What primary agent auto-onboarding does

When the primary agent is enabled, it uses auto-onboarding via a private channel along with a user greeting message.

Private channel behavior

The private channel created for onboarding is automatically deleted after 24 hours of inactivity.

Managing agents

After enabling and onboarding, you can manage the agents from the agent section.

Conclusion

To set up Spark Chat agents effectively, start with the required foundation—channels and response mode—then refine behavior using a custom prompt, conversation style, and response length. Next, add action rules for consistent logic (like spam alerts or routing users to resources), and finish with safety/compliance disclaimers and guard rails. If you want a guided start, enable the primary agent with auto-onboarding.