Configuring Agents

​

Overview

Once you’ve identified which proactive AI agents are right for your space, the next step is configuring each one to match your operational needs. The configuration panel lets you control when an agent acts, how often, through which channel, and what tone its messages use. This page walks you through every setting available in the Agent Configuration panel, from enabling an agent to testing voice calls. Route: Configure Agents (click any agent card to open its configuration)

---

​

Enabling an Agent

Before you can configure an agent, you need to enable it.

1. Navigate to Operations > AI Agents
2. Find the agent you want to enable in the grid
3. Click the toggle switch on the agent card
4. The status badge changes from Inactive (grey) to Active (green)

When you enable an agent for the first time, Nexudus automatically creates a configuration with default settings specific to that agent. These defaults are a safe starting point — you can customize them after enabling.

Tip: Enable one agent at a time. This lets you understand how each agent behaves before adding more.

---

​

Opening the Configuration Panel

After enabling an agent (or if it’s already enabled), click anywhere on the agent card to open its Configuration Panel. The panel is organized into three sections:

1. Basic Settings — Enable/disable and view agent description
2. Execution & Behavior — Auto-execute, confidence threshold, cooldown, and rate limits
3. Communication Settings — Channel, voice scripts, and custom instructions (outbound agents only)

At the bottom of the panel, you’ll also find a Test Configuration section for verifying voice call quality.

---

​

Basic Settings

​

Enable / Disable

The master switch for the agent. When disabled, the agent stops evaluating criteria and proposing actions immediately. Re-enabling it resumes monitoring from that point forward.

• Active (green badge) — Agent is monitoring and proposing actions
• Inactive (grey badge) — Agent is paused

Note: Disabling an agent does not delete your configuration. All your custom settings are preserved and will be applied when you re-enable it.

---

​

Execution & Behavior

These settings control when and how often the agent takes action. They appear once the agent is enabled.

​

Auto-Execution

Auto-execute allows high-confidence actions to be sent automatically without waiting for your manual approval in the AI Inbox. This is useful for routine, low-risk actions where you trust the AI’s judgment. How it works:

1. The agent detects a pattern and calculates a confidence score (0–100%)
2. If auto-execute is enabled and the score meets or exceeds your threshold, the action is sent immediately
3. Actions below the threshold still appear in your AI Inbox for manual review

When to enable auto-execute: Good candidates:

• Payment reminders (factual, expected communication)
• Contract renewal outreach (timely, relationship-building)
• Onboarding nudges (helpful, low-risk)

Keep manual review for:

• At-risk member detection (needs personal context)
• Support pattern outreach (requires understanding of issue severity)
• Any agent where the message tone needs your personal touch

Minimum Confidence Threshold When auto-execute is enabled, a slider appears allowing you to set the minimum confidence level required for automatic sending. Recommended approach:

1. Start with auto-execute disabled while you learn how the agent behaves
2. After 1–2 weeks of manual review, enable auto-execute at 85%
3. Adjust up or down based on how many actions you’re approving vs. rejecting

​

Rate Limiting

Rate limiting prevents your agents from over-communicating with customers. Two settings control this:

​

Cooldown Period (Hours)

The minimum number of hours that must pass before the same agent can act on the same customer again.

• Range: 1–720 hours (1 hour to 30 days)
• Default: Varies by agent (e.g., 48 hours for Due Invoice Reminders, 72 hours for Contract Renewal)

How it works: If the Due Invoice Reminders agent sends a reminder to a customer at 10:00 AM on Monday with a 48-hour cooldown, it won’t propose another action for that same customer until 10:00 AM on Wednesday — even if new patterns are detected. Recommended cooldowns by scenario:

​

Maximum Actions Per Day

The upper limit on how many actions this agent can propose across all customers in a single day.

• Range: 1–1,000
• Default: Varies by agent (e.g., 20 for Due Invoice Reminders, 10 for Contract Renewal)
• Leave empty for unlimited (not recommended for most agents)

Why set a limit? Without a cap, a single event (like a payment processing outage) could trigger hundreds of reminder actions at once. A daily limit ensures your communication stays controlled and prevents message fatigue. Recommended daily limits:

---

​

Communication Settings

These settings appear only for agents that send outbound messages (email or voice). Internal-only agents (like FAQ Gap Detection) skip this section entirely.

​

Preferred Channel

Override the default communication channel for this agent. When to use Voice:

• High-priority, urgent matters (e.g., tour reminders before an appointment)
• Personal touch for VIP or long-term customers
• After email attempts have not been acknowledged
• Requires voice channel configuration on your location

Note: Not all agents support voice calls. The channel badge on each agent card shows its default channel.

​

Available Placeholders

When configuring voice scripts, you’ll see a list of placeholders specific to this agent. These are dynamic variables that get replaced with actual customer data when the message is sent. To copy a placeholder, click on it. It will be copied to your clipboard and show a green checkmark briefly. Common placeholders include:

Tip: Each agent type has different available placeholders. The list shown in the configuration panel is specific to the agent you’re configuring.

​

Opening Message

Voice only — This setting applies exclusively to agents that send voice calls. It has no effect on email or other channels.

The first line spoken when the voice call is delivered. This overrides the AI’s default greeting.

• Maximum length: 500 characters
• Supports placeholders: Yes (e.g., {{company_name}}, {{customer_name}})

Example:

Hello {{customer_name}}, this is a quick call from {{company_name}}. I noticed your invoice is a few days overdue — is now a good time to speak?

​

System Prompt

Voice only — This setting applies exclusively to agents that send voice calls. It has no effect on email or other channels.

Full AI instructions for generating the voice message content. This gives you fine-grained control over tone, structure, and content.

• Maximum length: 2,000 characters
• Must include: The {{voice_message}} placeholder (required when using OVERRIDE mode)

Standard usage: Leave blank to use the agent’s default system prompt. The AI already knows the right tone and structure for each agent type. Advanced usage — Override mode: If you prefix your system prompt with OVERRIDE:, it replaces the entire default prompt. This is for advanced users who want complete control.

Important: When using OVERRIDE: in a voice agent’s system prompt, you must include the {{voice_message}} placeholder. If you don’t, the save will be rejected with an error.

Example override:

OVERRIDE: You are a friendly receptionist at a coworking space. Be warm, conversational, and brief. Never exceed 30 seconds of speaking time. Always end with a question that invites a response. {{voice_message}}

​

Custom Prompt Instructions

Additional instructions that are appended to the agent’s default prompt (rather than replacing it). This is the easiest way to customize message behavior across all agents — email and voice — without risking breaking the agent.

• Maximum length: 2,000 characters
• Applies to: All outbound agents (email and voice)
• Safe to use: Yes, for most operators
• Tip: Prefix with OVERRIDE: to replace the entire prompt instead of appending

Example:

Always address the customer by their first name. Use a friendly but professional tone. Never sound pushy about payments.

Best practices for custom prompts:

1. Start with small, specific instructions rather than rewriting the entire prompt
2. Test the result by reviewing approved actions before enabling auto-execute
3. Keep tone consistent with your location’s brand voice
4. Avoid conflicting instructions (e.g., “be brief” + “include full invoice details”)

---

​

Test Configuration

Once your configuration is saved, you can verify how it works in practice.

​

Test Voice Call

For agents that support voice calls, you can initiate a test call to hear exactly what your customers will hear.

1. Save your configuration first (unsaved changes prevent testing)
2. Enter your phone number in the Phone number field (include country code, e.g., +44123456789)
3. Click Test Voice Call
4. You’ll receive a phone call using your current agent configuration

What to verify during the test:

• Voice quality and clarity
• Opening message sounds natural
• Tone matches your location’s brand
• The call handles gracefully (you can hang up, respond to questions)
• Placeholders are replaced with real sample data

After the test call, you’ll see a success or error message indicating whether the call was initiated successfully.

---

​

Saving Your Configuration

Changes are saved by clicking the Save button in the bottom bar of the configuration panel. A green confirmation message appears briefly when the save is successful.

Important: If you have unsaved changes and try to initiate a test voice call, you’ll be prompted to save first. The test uses your saved configuration, not your in-progress changes.

---

​

Recommended Configuration Workflow

Follow this workflow when setting up a new agent:

​

Phase 1: Enable and Observe (Week 1–2)

1. Toggle the agent on
2. Leave auto-execute disabled
3. Use default cooldown and daily limits
4. Use default channel and prompts
5. Review proposed actions in your AI Inbox daily
6. Note patterns: which actions do you approve? Which do you reject or edit?

​

Phase 2: Tune (Week 3–4)

1. Adjust cooldown if you’re seeing too many or too few actions for the same customer
2. Adjust max actions per day if your inbox is overwhelmed or underutilized
3. If the agent is sending through the wrong channel, switch it
4. Tweak custom prompt instructions to improve tone or add specific details

​

Phase 3: Automate (Week 5+)

1. If your approval rate is high (80%+), consider enabling auto-execute
2. Set the confidence threshold to 85% initially
3. Monitor for 1–2 more weeks
4. Adjust the threshold based on the quality of auto-executed actions

---

​

Quick Reference: All Settings

---

​

What’s Next?

• Browse the Agent Types Reference for detailed information on each agent’s behavior and recommended settings
• Review and manage suggested actions in the AI Inbox
• View AI Conversations to see two-way conversations with customers