Blueprints

Upload videos to a storage folder. This automation polls the folder on a schedule, picks the oldest unprocessed clip, generates an engaging Facebook caption with Groq AI, and posts it to a specified Facebook Page using the Graph API. Uses the pages_manage_posts scope exclusively — no Page reads beyond confirming the post ID. Optional Telegram approval gate: before posting, the workflow sends a three-button message (Approve / Regenerate / Reject) with a caption preview, privacy setting, paid-partnership disclosure, and queue remaining count. Approve posts immediately. Regenerate creates a new AI caption with a fresh angle and asks again (capped at 2 retries). Reject marks the file processed without posting and stores rejection metadata. When the queue is empty, the workflow sends an informational Telegram message rather than silently no-op'ing — you always know when there is nothing to post.

Upload a CSV recipient list to the queue folder. Each CSV row is one recipient: columns `phone,name` (required) plus up to 4 optional template-parameter columns. This automation polls the folder, picks the oldest unprocessed CSV, parses it, and sends a pre-approved WhatsApp template message to every recipient — rate-limited to stay within Meta's per-number delivery ceilings. Approval model is different from posting blueprints: batch-level (one Approve/Cancel per campaign). No regeneration (templates are pre-approved in Meta Business Manager; parameters are per-recipient). Each send has its own success gate — failures are logged and reported in the summary, but don't abort the campaign. Rate limit: configurable sends-per-minute (default 10, max 60). Wait node enforces pacing between sends so you don't trip Meta's throttling. When the queue is empty, the workflow sends an informational Telegram message rather than silently no-op'ing.

When a customer sends a WhatsApp message to your registered Business Phone Number, the platform centralised webhook at /api/webhooks/whatsapp routes it to this automation (Architecture B — one Meta webhook URL, many per-tenant automations). **What V2 does differently from V1:** before generating a reply, it fetches the most recent N messages exchanged with that specific phone number from the platform's whatsapp_messages table. The AI sees the running conversation and stays coherent across multiple turns. After replying, it logs the outbound message back to the same table so the next inbound triggers a memory-aware reply. **Limits and caveats:** • Only operates on text messages in V1. Non-text (image, audio, video, document, interactive) gets a polite default reply. • Memory window is configurable (default last 10 messages) — keep small to stay within model context limits and reduce token cost. • Reply is free-form text, only valid within Meta's 24-hour customer-service window. If the user has been silent for >24h, the API will reject the send with code 131047 and the reply will be logged as failed. • No template fallback in V1. For outside-window engagement, use the Campaign Publisher blueprint with a pre-approved template instead.