> ## Documentation Index > Fetch the complete documentation index at: https://docs.guhan.ai/llms.txt > Use this file to discover all available pages before exploring further. # Sending safety > How Guhan keeps your LinkedIn account (and inboxes) safe. The fastest way to burn an outbound motion is to get your sender accounts restricted or spam-flagged. Guhan is built with safety as the default — not the setting you have to remember to turn on. ## LinkedIn safety LinkedIn watches for behavior that looks automated. Guhan mimics human patterns: ### Daily caps Every action type has a daily ceiling that stays inside LinkedIn's tolerance: | Action | Cap | Why | | -------------------------------- | -------- | ------------------------------------------------ | | Invites | 25/day | LinkedIn flags accounts sending 100+ invites/day | | DMs | 30/day | Standard human rate | | Profile views + lookups combined | 200/day | LinkedIn's soft ceiling on browsing volume | | Comments | 100/week | Weekly cap; comments cluster daily is fine | | Reactions | 300/week | Same | You can adjust these per sender, but Guhan warns you if you push past LinkedIn's published safe limits. ### Sending window Messages only go out during hours you set — default 10am–6pm in the sender's timezone. Weekends optional. Guhan **randomizes send times** within the window, spread across the day, so it doesn't look like a bot firing at exactly 10:00:00. ### Warm-up New sender accounts (\< 30 days) start at 30% of the daily cap and ramp over 30 days. Established accounts (existing history + connections) skip warm-up. ### Randomized inter-action delay Between two consecutive actions on the same account, Guhan waits a randomized 60-240 seconds. No bursts, no evenly-spaced 1-per-minute pattern. ### Auto-withdraw stale invites Pending invites older than 14 days get auto-withdrawn. LinkedIn dislikes stacks of unaccepted invites on an account — auto-withdraw keeps the pending queue clean. ### Never comment/react without content Guhan requires the prospect to have a recent public post before scheduling a comment or reaction step. No content = the step gets skipped, no fake "reacted to their profile" behavior. ## Email safety For Gmail / Outlook: * **Warm-up** — new email senders start at 30% capacity, ramp over 30 days * **Reply tracking** — replies land back in your Gmail thread + in Guhan's Inbox; no forwarding, no header manipulation * **Unsubscribe honored** — anyone who replies "unsubscribe" or similar goes on your workspace-level Do Not Contact list * **No BCC blasts** — every email is a real 1:1 thread; no mailing-list headers ## WhatsApp safety Personal / business WhatsApp accounts have low daily caps by design: * Default 40 messages/day per account * No template-message bulk-sending — every message is generated per prospect * Reply capture is real-time ## Do Not Contact You can add: * **Specific email addresses** — never message this person * **Specific LinkedIn URLs** — never message this person * **Domains** — never message anyone at this company (e.g., your own domain, competitor domains) * **Prospects who reply "not interested"** — auto-added The DNC list is enforced globally across your workspace — every Outreach Agent + every Watchlist Agent respects it. ## Health monitoring The **Dashboard** and **Settings → Sender accounts** page show: * **Green** — sender is healthy, sending normally * **Amber** — sender needs attention (approaching daily cap, near sending-window end) * **Red** — sender is restricted or needs re-authentication Reauth flows are one-click — re-sign in with LinkedIn/Google/Microsoft and you're back in business. ## What happens if you push too hard If you widen daily caps beyond LinkedIn's safe zone or shorten the sending window aggressively, Guhan will show warnings but let you do it (it's your account). If you get restricted: * Guhan pauses that sender * The banner tells you what happened * Reauth once LinkedIn lifts the restriction * Guhan resumes at reduced pace to rebuild reputation ## Related Daily limits and warm-up detail. Fixing a broken sender.