Labeeq for iPhone and Android is coming soon. Join the waitlist

العربية Sign in Start free trial

Retail & e-commerce

Answer product questions, capture purchase intent and follow up on the channels your shoppers use.

Product guidance from maintained knowledge

Purchase requests and follow-up in one thread

Consent-based offers to eligible segments

See how it fits

From the blog WhatsApp Business for Companies: Complete Guide Read the article

AI agents

Configure AI agents for customer conversations

An active, assigned agent can draft customer replies from its instructions and selected knowledge, use enabled workspace actions, and hand a conversation to a named teammate, team or AI agent.

Start free trial See pricing

Use maintained knowledge

Attach selected text, supported files or crawled website sources, then add instructions and example exchanges. Keep those sources current and decide whether general model knowledge is allowed.

Assignment is explicit

The agent responds while it is active and assigned. With assignment enabled, it can transfer the thread to an available person, team or another agent; the existing conversation history remains visible.

Actions have controls

Enable actions such as assignment, memory, existing labels, lifecycle changes, internal notes, flow starts and closing. Contact-field capture is enabled by default and can be switched off.

How it works

1
Create an agent, choose its language behavior, and write clear scope and escalation instructions.
2
Attach the relevant knowledge sources, examples and saved media, then enable only the actions it needs.
3
Use the playground and test conversations for known, unknown, sensitive and handoff cases.
4
Assign it manually or through a flow, then review real replies, actions, handoffs and AI allowance.

Best fit

Teams with repeatable customer questions, maintained source material and a clear human escalation path.

Set up deliberately

Define what the agent may answer, what it must never decide, and exactly who receives each escalation.
Select only relevant knowledge, configure language and actions, and test adversarial as well as ordinary questions.
Monitor provider availability, monthly credits, action logs and resolved conversations; revise instructions and sources as facts change.

Confirm before launch

AI output is probabilistic: grounding and instructions reduce mistakes but do not guarantee factual, compliant or dialect-perfect answers.
Replies require a configured AI provider and available monthly credits or an add-on; delivery still depends on the connected channel.
Conversation content may be processed by the configured model provider. Review your provider contract, data controls and legal obligations before sending sensitive data.

Bot flows vs an AI agent

Labeeq has both — and they do different jobs. Flows follow the branches you draw; the agent holds a conversation.

	Bot flows	AI agent
Best for	Predictable journeys — menus, routing, collecting details	Open questions asked in natural language
Input	Defined triggers, buttons and saved answers	Natural-language customer messages
Behaviour	Runs configured nodes, conditions and actions	Generates replies and proposes enabled actions
Unknown requests	Follow the fallback path you configured	May answer or use a configured handoff path
Setup	Drag-and-drop builder	Instructions + knowledge documents
Cost	Included in every plan	Free monthly credits; add-on packages beyond

They can work together: a flow handles deterministic routing, then an assigned agent handles suitable free-text questions until a configured handoff applies.

Controls available to your team

Replies while assigned

An active agent can handle eligible inbound turns outside staffed hours as well as during the day, subject to provider availability, credits and channel delivery.

Configurable language behavior

Choose a fixed language or instruct the agent to follow the customer's latest language. The rule guides the model; production testing confirms whether its wording suits your audience.

Actions beyond text

Per-agent controls cover assignment, memory, existing labels, lifecycle changes, notes, flows and closure; contact-field capture starts enabled and remains configurable.

Handoffs with optional context

The assignment action can transfer to an available named target. When internal notes are also enabled, the generated handoff can include a private context note.

Optional timed behavior

Instructions can compile into reviewed follow-up or close steps for quiet conversations. Keep the agent assigned and test timing, message content and cancellation behavior.

Named agent-to-agent routing

An agent can assign to another available AI agent by exact name when the assignment action and target are configured; test loops and fallback ownership before launch.

Safeguards to configure and verify

Grounding controls

Select maintained business sources and decide whether general knowledge is allowed. Grounding reduces unsupported answers but still needs monitoring and escalation tests.

Permission-gated actions

Most workspace actions are opt-in for each agent and can carry operator guidance. Contact-field capture is enabled by default, so review that toggle too.

Provider-aware data review

The configured provider processes conversation context to generate output. Confirm its retention, training and regional terms against your own data requirements.

Active and assigned state

Automated replies require the agent to remain active and assigned. Turn it off or reassign future handling to a person, then confirm the thread's resulting ownership.

AI agents

Common questions

Which languages can the agent use?

You can fix an agent to Arabic or English, choose another configured language, or instruct it to match the customer's latest language. The model generates the wording, so test the languages and dialects your customers actually use.

What does the AI cost?

Each plan currently includes a monthly AI-credit allowance. When it is exhausted, AI responses pause unless you add an AI package; allowances reset by calendar month.

Can it hand a conversation to a human?

Yes, when the assignment action is enabled and a valid person or team is available. Enable internal notes as well if you want the agent to leave its generated handoff context for the team.

What can it do besides replying?

Available controls include capturing contact fields, remembering durable facts, assigning, adding existing labels, changing the contact lifecycle, leaving notes, starting an active flow, sending saved media and closing a resolved thread. Enable and test each one deliberately.

How do I teach it my business?

Write instructions and examples, then attach selected plain text, supported text or PDF files, or website knowledge sources. Review crawled content and resync sources when your catalogue or policies change.

Will it ever make a wrong answer?

It can. Restricting it to maintained knowledge, disabling general knowledge and defining escalation rules reduces risk but cannot eliminate model errors. Test regularly and reserve high-impact decisions for people.

How fast can it go live?

There is no safe universal time estimate. Readiness depends on provider configuration, clean source material, action and routing setup, and successful testing of normal, unknown and sensitive requests.

Can it cover only nights and weekends?

Yes, if a tested flow assigns eligible after-hours conversations to the active agent and routes daytime or escalation cases to your team. Business-hours and fallback rules must be configured.