> ## Documentation Index > Fetch the complete documentation index at: https://docs.kapso.ai/llms.txt > Use this file to discover all available pages before exploring further. # Messaging > Send and receive messages from the inbox ## When you can send messages You can send messages when: * The conversation is **active** (not ended) * No workflow is running, or the workflow is in **handoff** or **failed** state If a workflow is running, the input is disabled. Click **Handoff** to take control. See [Automation and handoff](/docs/platform/inbox/automation). ## Message types | Type | Max size | Notes | | ------------- | -------- | ---------------------------------------- | | Text | — | Enter to send, Shift+Enter for newlines | | Image | 5 MB | Via file picker, drag-and-drop, or paste | | Video | 16 MB | | | Audio | 16 MB | | | Document | 100 MB | | | Voice message | — | Record from browser mic | You can also reply to specific messages (quoted replies). Reactions from contacts are shown as emoji badges on the target message. ## Interactive messages Send interactive WhatsApp messages via the `+` attachment menu: | Type | Details | | ------------- | --------------------------------------------------------------------------------------------------------- | | Buttons | Body + up to 3 reply buttons (max 20 chars each). Optional header (text/image/video/document) and footer. | | List | Body + sections with up to 10 total rows. Each row has a title and optional description. | | CTA URL | Body + button with text and URL. Optional header and footer. | | WhatsApp Flow | Select a published Flow, set body, header, footer, and CTA button text. | ## Quick replies Saved message templates you insert into the composer with a `/` shortcut. Type `/` followed by the shortcut in the message input. The picker filters as you type — shortcut prefix matches rank first, then name prefix, then any substring. Select one and its body replaces the `/shortcut` token. Manage quick replies from **Quick replies** in the sidebar. Each one has: * **Name** — up to 120 characters * **Shortcut** — up to 40 characters, lowercase letters, numbers, hyphens, and underscores only * **Body** — up to 10,000 characters * **Visibility** — `personal` (only you) or `shared` (everyone in the project) Shortcuts are unique per project. ### Variables Bodies can include variables that resolve against the conversation's contact when inserted. Type `{{` in the body editor to pick one. | Variable | Resolves to | | -------------------------- | ---------------------------------------- | | `{{contact.first_name}}` | First word of the contact's display name | | `{{contact.name}}` | Contact's display name | | `{{contact.phone_number}}` | Contact's phone number | If the contact has no display name, `contact.first_name` and `contact.name` fall back to the phone number. Any other variable is rejected when saving. ### AI prompts Bodies can also include prompts that an AI fills in when the reply is inserted. Type `{{` in the body editor and pick **AI prompt**, or write the syntax directly: ```text theme={null} Your refund ID is {{prompt: "Find the refund ID in this conversation"}}. ``` The prompt text is a JSON string — use `\"` for a literal quote. When you insert the reply, each prompt section shows a generating placeholder in the composer and is replaced with the generated text. Sending and voice recording are blocked while generation is running. What the AI can use: * The recent messages of the current conversation * Read-only inbox lookups (listing conversations, reading conversation messages) * Public web search, for current or public facts only — never for customer-specific data such as orders, refunds, or account details Generated text is written to fit the surrounding sentence: it matches the capitalization, spacing, and punctuation around the slot, and the text outside the slot is never changed. If essential context is missing, the AI returns a short clarification request instead of inventing a value. Limits: * Up to 10 prompts per reply * Up to 2,000 characters per prompt * Prompt text cannot be empty * Malformed prompt syntax is rejected when saving Resolution consumes AI credits from the project. In the embedded inbox, resolution is limited to 10 requests per minute per inbox token. ### Permissions | Action | Who | | -------------------------------------- | ------------------------- | | Create, edit, archive a personal reply | Its author | | Create, edit, archive a shared reply | Project owners and admins | | Change visibility | Project owners and admins | | Duplicate any visible reply | Any project member | Duplicating creates a personal copy owned by you, named ` copy` with a `-copy` shortcut suffix. Archiving hides a reply from the picker without deleting it. Use the **Archived** filter on the management page to restore it. In the [embedded inbox](/docs/platform/inbox/embedded), the `/` picker works but quick replies cannot be managed. ## Starting a new conversation Click the compose button in the conversation list header to start a new conversation: 1. Select a WhatsApp number (production numbers only) 2. Enter the recipient's phone number with country code 3. Pick an approved template 4. Fill in template parameters 5. Send ## Delivery errors Failed messages show a red warning icon with the error from WhatsApp. Common causes: * 24-hour customer service window expired * Template not approved * Invalid phone number ## Assignments Assign conversations to team members from the Info tab in the detail sidebar. * One active assignment per conversation * Creating a new assignment automatically deactivates the previous one * Not available in the [embedded inbox](/docs/platform/inbox/embedded) ## Contact info The Info tab in the detail sidebar shows: * **Display name** — Editable inline * **Phone number** * **Metadata** — Read-only, set via API or workflows * **Notes** — Add and edit notes on a contact (session auth only, not available in embedded inbox) ## Conversation history The Info tab shows previous conversations with the same phone number: date range, message count, and a link to view the full thread.