AI Agent (Persona)

Your AI agent is the always-on teammate that responds to customer messages, answers questions from what you've taught it, and hands off to a human when needed.

Use it as your first line of support so your team can focus on conversations that really need a human touch.

What your AI agent can do

• Reply automatically — Customers get instant responses, even at 3am
• Use your Knowledge Base — Pulls from the info you've added
• Answer common questions — Handles FAQs without you lifting a finger
• Remember context — Tracks the conversation within each chat
• Handle volume — Runs ten conversations at once without breaking a sweat
• Know its limits — Hands off to your team when it's out of its depth

Set up your AI agent

Open AI Persona settings

1. From your dashboard, go to Intelligence → Agent.
2. Select an existing AI Persona or create a new one.

The five configuration tabs

Each AI Persona is configured across five tabs:

1. Identity — Name, personality, languages
2. Instructions — How it should behave and respond
3. Skills — Actions it can take
4. CSAT — Satisfaction surveys for this agent
5. Advanced Settings — Fine-tuning for specific needs

Configure each tab

Tab 1: Identity

Define who your agent is.

Give it a name and greeting

• Nickname — Something friendly like "Alex" or "Support Team"
• Description — A short note about what your agent does
• Greeting — The first message customers see when they start chatting

Languages

The agent understands messages in any language, but you choose which languages it responds in. Pick the ones your customers actually speak.

Available options: English, Arabic, Hindi, Urdu, Spanish, French, German, Malayalam

Follow-up messages

When a customer goes quiet, your agent can check in automatically. Choose how long to wait before sending the follow-up:

• No follow-up — Don't send one
• 1 minute — Quick nudge
• 10 minutes — Short pause
• 1 hour — Give them some time
• Custom — Anywhere from 2 to 23 hours

Follow-ups re-engage people who got distracted or missed the notification.

Response length

How detailed should answers be?

• Tiny — Very short
• Concise — Quick, to the point
• Medium — Balanced
• Long — Thorough and detailed

Personality

How should the agent come across?

• Professional — Formal, business-like
• Friendly — Warm and approachable
• Casual — Relaxed and conversational

Pick what matches your brand.

Handover to humans

Sometimes AI isn't the right fit. Add custom handover rules to tell your agent when to step aside and bring in a real person.

Click + to add a condition, then describe the scenario in plain language — for example, "Customer asks for a refund" or "Customer requests to speak to a manager." Add, edit, and remove conditions anytime.

When any condition matches during a conversation, the agent hands off. Built-in triggers also cover cases like an explicit request for a human, frustration, sensitive issues, and other escalations — your custom conditions sit alongside those.

Automatic assignment to a team

On the Starter plan and above, if you have at least one team with its team inbox enabled, the AI can route the handover into the best-matching team instead of only the general queue.

How it works:

1. The agent decides a handover is needed (custom or built-in conditions).
2. Cloodot matches the handover reason against each inbox-enabled team's description.
3. If a team clearly fits, the conversation is moved into that team inbox and the AI assignee is cleared so the team owns it.
4. If no team clearly fits — or team inboxes aren't available — the conversation goes to Queued for the next available agent.

This is different from routing a conversation to a team by hand in the inbox: manual routing keeps the current agent; AI handover clears the assignee and transfers ownership to the team (or the queue).

To get accurate auto-routing:

• Turn on Team inbox for the teams that should receive escalations (Teams)
• Write a clear description for each of those teams (what they handle, not just a name)
• Keep handover conditions specific so the reason the AI records is useful for matching

See Inbox for how team inboxes appear in the sidebar, and Teams for setup.

Tab 2: Instructions

Tell the agent how to behave. Think of it as training a new teammate — you're explaining how you want them to handle different situations.

What to include

• General guidelines — How to greet people, what tone to use
• Specific scenarios — "If someone asks about returns, explain our 30-day policy"
• Things to avoid — "Don't promise specific delivery dates"

Organize with instruction blocks

Break instructions into instruction blocks for different scenarios — especially when you want quick reply buttons or a specific skill to run at the right moment.

Each block has:

• A description — when the block should apply
• Optional positive and negative example messages — for sharper matching
• Instructions — what the agent should do in that situation
• Quick replies and skill references — interactive options for customers

See Instruction Blocks for the full setup guide, and Writing Block Descriptions for help with descriptions and examples.

Tab 3: Skills

Skills let your agent actually do things, not just answer questions.

What you'll see

• Default skills — Built-in abilities every agent has
• Installed skills — Custom ones you've added

What skills can do

• Check order status
• Book appointments
• Look up information
• Connect to your other tools

Tab 4: CSAT

Turn on customer-satisfaction surveys for this agent. Customers can then rate the conversations it handles, and the results feed into your CSAT dashboard.

Tab 5: Advanced Settings

Fine-tune how the agent behaves.

Temperature

Controls how creative vs. predictable the agent is. The slider has labeled stops — strict (0.25), neutral (0.5), and creative (0.75):

• Lower (toward strict) — More consistent, predictable responses
• Higher (toward creative) — More varied, creative responses

Default is 0.75, which works well for most cases. Lower it for factual or sensitive topics that need more predictable answers.

Experimental AI

Toggle on to enable experimental AI features for improved responses. The setting applies per persona, so you can try it on one agent without affecting the rest.

Fallback message

When the AI gets stuck, it sends this message instead of leaving the customer hanging. Write something helpful — apologize and offer to connect them with a real person.

Save, test, and deploy

Save changes

Click Save after editing. Each save creates a new version, so you can always roll back if something doesn't work.

Versions can also be saved as drafts from external tools — for example, via the Cloodot MCP server or the Developer API — so you can prepare changes outside the dashboard. Saves from those tools always create a draft; they never replace what's live.

Test in the playground

Click Play to open the playground. Try some conversations, see how the agent responds, and tweak until you're happy.

Deploy

When you're happy with how the agent is working, deploy it. That version starts handling customer conversations. Update and redeploy anytime.

After you deploy

What the agent handles automatically

• Responds to new customer messages
• Searches your Knowledge Base for relevant info
• Uses FAQs to answer common questions
• Follows your instructions
• Remembers context within each conversation

Monitor performance

Check in on how the agent is doing:

• How many conversations it's handling
• Response quality
• When it hands off to humans
• Customer satisfaction
• Which skills get used most

Message credits and AI assignment

The AI agent needs incoming message credits to respond. When credits run out:

• New conversations — The agent isn't assigned
• Existing conversations — If the agent was previously assigned and a customer messages back, it's automatically unassigned since it can't respond without credits

To check remaining credits, go to Settings → Billing.

Version history

Every save creates a new version, so you can:

• See what changed over time
• Roll back to a previous version if something breaks
• Test new ideas without risking your working setup

Tips for a great agent

Identity

• Pick a name that fits your brand
• Write a greeting that's welcoming and sets expectations
• Choose languages your customers actually speak
• Match the personality to your brand voice

Instructions

• Be specific — vague instructions lead to vague answers
• Group related scenarios into instruction blocks
• Write clear block descriptions and add quick replies where they help
• Update based on real conversations
• Test in the playground before deploying

Skills

• Install only what you actually need
• Configure each skill correctly

Advanced settings

• Start with the default temperature; adjust only if needed
• Write a fallback message that helps, not one that frustrates

Handover and teams

• Write specific handover conditions so escalations fire for the right reasons
• Give each inbox-enabled team a clear description of what it handles — the AI uses that to pick a team on handover
• Only enable team inboxes for teams that should receive escalations

Overall

• Test before you deploy — The playground is there for a reason
• Keep improving — Review conversations and refine your setup
• Build your Knowledge Base — Better info means better answers
• Keep FAQs current — Outdated answers are worse than none

Troubleshooting

Agent not responding

Check that an active version is deployed. Then verify:

• The channels are connected
• The agent is assigned to handle conversations

Answers aren't great

The agent usually needs more information. Try:

• Adding more to your Knowledge Base
• Updating your FAQs
• Making your instructions more specific
• Adjusting the temperature setting

Too many handoffs

The agent may not have enough info to answer confidently. Try:

• Filling gaps in your Knowledge Base
• Adding FAQs for common questions
• Reviewing which questions are causing handoffs

Extend your agent

Add capabilities

• Skill Sets — Add custom abilities
• Creating Skills — Build your own
• Skill Schema — Technical reference

Connect more

• MCP Servers — Connect external tools
• Knowledge Base — Teach your agent more
• Collections — Showcase products and services in conversations
• FAQs — Handle common questions
• Teams — Enable team inboxes for automatic handover routing
• Inbox — Work conversations after a handover lands in a team or the queue

What's next?

1. Set up instruction blocks — Add quick replies and skills for interactive scenarios
2. Build your Knowledge Base — Give your agent the info it needs
3. Create FAQs — Cover the questions you hear most
4. Explore Skill Sets — Add custom capabilities
5. Test in the playground until you're happy

---

Ready to teach your agent? Add to your Knowledge Base.