Write for every channel. In one click.

What the AI can write

Channel-aware generation

When you generate a caption from the compose page, you select which channel it is for. The AI receives that channel's real limits — character cap, hashtag cap, media constraints — and writes within them. You never have to manually trim a Twitter caption that went over 280, or count hashtags for Instagram.

• X / Twitter: 280-character cap respected, up to 3 hashtags
• Instagram: up to 2,200 characters, up to 30 hashtags
• LinkedIn: professional tone, up to 3,000 characters, up to 5 hashtags
• Threads: concise, up to 500 characters
• Every other channel follows its own platform rules

Copywriting frameworks

Pick a framework from the compose page and the AI uses it to structure your caption:

• AIDA — Attention, Interest, Desire, Action. Classic structure that moves the reader toward a click or a purchase.
• PAS — Problem, Agitate, Solution. Leads with a pain point, then resolves it with your offer.

Set a default framework in your AI settings — every generated caption will follow it unless you override it on a specific post. Your team's brand voice stays consistent without having to remember to re-select it each time.

WhatsApp template generation

WhatsApp templates have strict rules: category-specific structure, no prohibited content, variable placeholders in the right format, character limits on header, body and footer. The AI assistant builds templates that follow Meta's requirements automatically — and validates the output before you ever see it.

• Generates header, body and footer together as one structured template
• Enforces Meta's rules for Marketing, Utility and Authentication categories
• Suggests which contact merge fields to use as variables (e.g. first name, order number)
• If the first attempt breaks a rule, it retries automatically with the specific failure fed back into the prompt — up to two extra attempts before surfacing anything to you

The full prompt — including Meta's rules and compliance constraints — is assembled server-side. Nothing about those rules is sent from the browser, so the AI cannot be prompted by a user to bypass them.

Rate limits and safety

The AI is a writing assistant — it drafts, you review. A few guardrails keep it that way:

• 30 requests per account per hour, 10 per user per minute — enough for heavy daily use, limited enough to prevent automated abuse
• Your input is sanitised and capped at 500 characters before it reaches the model
• An anti-injection instruction on every prompt tells the model to ignore instructions embedded in your content input
• Nothing goes live until you review and click Publish — the AI never triggers publishing

Model fallback — it keeps working even when a model goes down

The AI runs on a primary model with an automatic fallback. If the primary model returns an error or takes too long, the request is retried on the fallback model transparently — you see the same result either way. After three consecutive failures, the primary is skipped for five minutes while the fallback handles traffic, then probed again automatically.