LinkedBot

Keeping a professional presence going on LinkedIn is mostly small writing jobs — a reply to someone who reached out, a note attached to a connection request, a short post about something that actually happened this month. Each one is minor; together they are the reason the account goes quiet for six weeks at a time. Most tools built to help with this automate the wrong half. They send at volume and leave the judgement to a template, which is how the platform filled up with messages anyone can spot within a line and a half. The part that genuinely takes time — working out what to say to this particular person, in your own voice — is the part bulk automation never touches.

LinkedBot inverts that split. It drafts; you decide. Messages, replies and posts arrive in a confirmation console as drafts, each with its revision history, and nothing leaves the machine until it is approved there by hand. A draft can go back for another pass as many times as it takes. The human step is not a disclaimer bolted on afterwards — the draft state machine is built around it, and there is no mode that skips it.

It runs entirely on the machine it is installed on. Contacts, drafts and history stay in local storage, and the AI capability comes from Local Agent Gateway running on the same machine rather than from a service somewhere else. LinkedBot started in May 2026 as a set of parts a developer had to assemble — a browser extension, a local daemon, and a bridge between them. Since August 2026 it has been the first app converted to run inside the runtime PrimeForge Desktop provides, which turns installing it into a download rather than a setup: the parts arrive as one package instead of three things to wire together. Most of the work since the first version has gone into removing the requirement to be technical.

LinkedBot suits one person keeping up a professional presence that matters — a founder, a consultant, anyone for whom LinkedIn is part of how work arrives — who writes carefully and wants help with the first draft rather than a machine speaking on their behalf. It suits people who would rather their contacts and their message history sat on their own machine than in another service's database. And it suits anyone already tired of automated outreach who wants a tool whose design makes their own judgement mandatory rather than optional.

It does not suit bulk outreach. There are no mass connection requests, no sequences fired at a purchased list, and no unattended sending — that is a design decision rather than a missing feature, and it is not on the roadmap. It does not suit a team wanting one dashboard across many accounts: this is one person, one machine, one account. And because it is labelled Evolving, individual screens and steps can still change between versions.

There is no public download page yet — access starts with a message through the contact page, which DarkHorseOne's engineering team reads and usually answers within two working days.