Deployment Channels: Put Your Agent on Web, WhatsApp, or Internal Tools

You can deploy an agent on:

1. Web — as a chat widget, embedded component, or standalone page. Best for customer-facing interactions, support, and lead capture.
2. WhatsApp / SMS — as a messaging bot. Best for notifications, quick Q&A, and reach in mobile-first markets.
3. Internal tools — inside Slack, Teams, or ticketing systems. Best for team workflows, approvals, and operational coordination.

Most agents start on one channel and expand. Pick the channel where your users already spend time.

Web deployment gives you the most control over the user experience:

• Chat widget — a floating button that opens a conversation panel. Good for support and lead capture on existing websites.
• Embedded component — the agent is integrated into a specific page or workflow. Good for onboarding, intake forms, and guided processes.
• Standalone page — a dedicated page for the agent, like ind.ai/yourname. Good for public-facing agent profiles.

Key requirements:

• Authentication — decide who can access the agent: anyone (public), logged-in users, or specific roles.
• Session management — maintain conversation context across page loads and browser sessions.
• Bot disclosure — clearly indicate that the user is interacting with an AI agent, not a human.
• Mobile responsiveness — ensure the widget works on all screen sizes.

WhatsApp is powerful for reach but has strict constraints:

• Opt-in consent — users must explicitly agree to receive messages from your agent. No cold outreach.
• Message templates — outbound messages must use pre-approved templates. Free-form messages are only allowed within a 24-hour window after the user's last message.
• Character limits — messages should be concise. Long responses should be split or linked to a web page.
• Media support — WhatsApp supports images, documents, and buttons, but each has formatting constraints.

Onboarding flow:

1. User sends a keyword (e.g., "START") or scans a QR code.
2. Agent responds with a welcome message and consent request.
3. User confirms consent.
4. Agent begins interaction within its defined scope.

Always provide a clear way to opt out (e.g., "STOP" keyword).

Internal deployment is often the fastest path to value:

• Slack / Teams bots — agents that respond in channels or DMs. Good for quick lookups, approvals, and status updates.
• Ticketing integration — agents that triage, categorise, and route tickets. Good for support teams and operations.
• Dashboard agents — agents embedded in internal dashboards that answer questions about data. Good for analytics and reporting.

Key requirements:

• Role-based access — the agent should respect your organisation's permission model.
• Thread context — maintain conversation context within Slack threads or Teams channels.
• Workflow integration — trigger agent actions from existing workflows.
• Notification management — don't spam channels. Use threads, DMs, or digest summaries.

Regardless of channel, every deployed agent needs:

• Human handover flow — a clear path for the agent to transfer to a human when it can't handle a request, including conversation context.
• Uptime monitoring — alerts when the agent is down or responding slowly.
• Usage analytics — track conversations, resolution rates, escalation rates, and user satisfaction.
• Content updates — a process for updating the agent's knowledge base, prompts, and responses as your business changes.
• Compliance — data handling, privacy policies, and disclosure requirements vary by channel and jurisdiction.

Each channel has unique risks:

Web:

• Rate limit conversations per IP to prevent abuse.
• Sanitise all user inputs to prevent injection attacks.
• Log all interactions for audit and improvement.

WhatsApp:

• Respect message windows—don't send unsolicited messages outside the 24-hour window.
• Keep responses under character limits to avoid truncation.
• Provide clear opt-out at every interaction.

Internal:

• Scope agent access to relevant channels and data.
• Avoid @everyone or @channel notifications from agents.
• Log all actions for internal audit and compliance.

Before going live on any channel:

• ☐ Bot disclosure is visible — users know they're talking to an AI.
• ☐ Human handover works — tested end-to-end, context is preserved.
• ☐ Consent flow is complete — users opt in before the agent engages (especially WhatsApp).
• ☐ Opt-out mechanism exists — users can stop interactions at any time.
• ☐ Rate limits are active — per-user and per-channel caps in place.
• ☐ Logging is on — every interaction recorded for audit.
• ☐ Fallback responses work — the agent handles unknown inputs gracefully.
• ☐ Mobile tested — widget/messages render correctly on all devices.
• ☐ Stakeholders briefed — team knows the agent is live and how to escalate issues.
• ☐ Monitoring dashboard active — real-time view of conversations, errors, and escalations.