LinkedIn

Runs in your browser

A Manifest V3 extension paired to your workspace. Its service worker polls the Kernel about every 75 seconds, picks up one action, has a content script run it by calling LinkedIn's internal Voyager API the way the site does, then reports back the result. One action per cycle, at human pace. A device carries one LinkedIn identity but can serve several workspaces: your multi-brand case, or an agency with one browser per client.

• Actions: invite, message, comment, reaction, visit, profile scrape
• Post publishing, search-result import, reactions and feed scraping
• connect, visit and scrape_profile are the most stable; message, comment, react are best-effort
• Developer-mode install for now, unpackaged extension

Anti-ban policy

The heart of the module. It decides what a device may do, right now. Hard per-action, per-identity ceilings can never be exceeded, even by an over-set config, because that is how an account gets banned. The daily count is a rolling 24-hour window, with no timezone arithmetic.

• Hard daily ceilings: 30 invites, 40 messages, 25 comments, 60 reactions
• Per-device sending window, in its own timezone
• 3-day ramp-up for a freshly paired account (35 then 55 then 75 %)
• Randomized 45 to 180 second gaps between actions, batches of 3 max per poll
• Per-device circuit breaker that pauses then auto-resumes

Multi-step outreach sequences

The direct Waalaxy replacement. A sequence chains steps: profile visit to warm up, invitation, message. You enroll contacts, a state machine walks them forward over time, each step goes through the queue and therefore the caps. The sequence pauses as soon as the prospect replies. Per-sequence daily cap on top of the device ceilings.

Closer bot on inbound DMs

When a prospect replies, the closer can draft the follow-up in your voice. Three modes: off (default), draft (it suggests, you approve before sending), send (it queues the reply straight as a normal message job, so under the same caps). The prompt carries the workspace's offer name, promise and CTA. The output is stripped of em-dashes and AI tropes before it is offered.

Automatic feed engagement

Optional, off by default. Three modes: off, like, like plus a short AI comment. The module selects relevant feed posts, bounds the number of likes and comments per pass, and never engages an excluded author. In parallel, a pass scrapes the reactions on your own published posts so reactors and commenters drop into the CRM.

Platform health and action queue

Distinct from the per-account breaker: when the same action fails at scale across several devices, that signals LinkedIn changed its API. A kill-switch disables that action fleet-wide, re-probes it after a cooldown, disables it again if still broken. A reaper frees jobs stuck mid-run. A per-action table shows successes, failures, pending and success rate over 7 days.