Security. What the code does today.

How Yappily protects your customers' data and keeps agents inside the rules you set. Where something isn't done yet, this page says so.

Only people approve

Approving a draft is person-only: an agent that tries through the API, MCP or the CLI gets a 403. New workspaces start in dry run. An agent's reply sends without a person only when its category is set to Auto in the app, the workspace is out of dry run and the agent's key may auto-send. New keys can't. Agent replies on Facebook and Instagram always stay drafts.

API keys

A key is shown once, and we store only its SHA-256 hash. Each key belongs to one workspace, taken from the key and never from the request. Revoking a key takes effect at once. Each key is limited to 120 requests a minute.

Workspace boundaries

Before reading or writing customer data, every request finds its workspace through sign-in, a widget key, an API key or the channel connection the message came in on.

Signed webhooks out

Deliveries are signed with HMAC-SHA256 to the Standard Webhooks spec. Endpoints must be public HTTPS; private and internal addresses are refused on every attempt, and redirects aren't followed. Every attempt is logged.

Verified webhooks in

Messages and events from Meta, Shopify, Stripe, Slack, Telegram and email are checked against their signature or secret before Yappily acts on them.

Channel tokens

Facebook and Instagram access tokens are encrypted with AES-256-GCM and bound to their workspace. Other channel tokens, such as Shopify, Slack and Telegram, aren't encrypted this way yet.

Knowledge crawling

When Yappily fetches a page for a knowledge base, it refuses private and internal addresses and checks each redirect again.

Staff access

Yappily's platform admin needs a role granted to a named account. It shows workspace names and dry-run state, and gives no way to act as a customer or read channel tokens.

Found a problem, or need an answer for a security review? Contact us.

Give your agents one record to work through.

Connect your brands and channels, point your agents at the API, MCP or CLI, and choose which kinds of reply wait for your team.