TL;DR: With this free plugin, you can delegate work to other harnesses and models, while keeping your main session the orchestrator. Nothing new to learn. Keep working as you do, but better. --- Remember a time when $20 a month bought one AI subscription, and it was enough? Ah, the good old days… Most of us run more than one now, and plenty of us pay a $100 minimum across Codex, Claude Code, etc. But we rarely get our money's worth out of any of them. It usually goes like this: You settle into one platform and one way of working, and you stay there until you hit a limit and switch to another tool, and it feels like starting over. New app, new tab, you reexplain the same context again as your time drains and your tokens with it. Then you find yourself at 90% weekly limit on Claude, but 50% unused and about-to-renew on Codex, or vice versa. I wanted to keep working the way I already do, and call in external models only when the leading session might need one. Sure, sometimes we tell Claude to call Codex or vice versa. But it rarely runs smoothly; it ignores your limits and costs, and hey, why box ourselves in? There's also a wave of strong open + free models now (GLM 5.2, HY3, etc). I wanted to hand work to those too, through OpenRouter or locally. ***So I built the most charming sorcerer your agent will ever meet: 🧙 OUTSOURCERER!*** 🧙 Outsourcerer turns any session into an outsourcing project manager. 🪄 It auto-detects what you already have - your existing subscriptions sitting there collecting dust. Then routes the work to the right subs and models at the right effort level while making sure they actually deliver, and optimizing cost + efficiency. ***Oh, and there’s literally no new tool to learn, no new commands, no configs to set up. Simply TALK to it in the session you already run.*** → Need extra intelligence for Opus? Better call Sol via Codex, or HY3 via OpenRouter, or Fable through another subscription you already have (say, Devin). → Want a council of wise advisors with Fable, HY3, Sol? The sorcerer runs it for you.

Quick context so the question makes sense: I run LinkedIn lead gen (I book qualified sales calls). Right now the way I find prospects is fully manual. I follow a handful of "watering hole" creators in my niche, people whose audience is basically my ideal client. Every time one of them posts something good, I go through everyone who liked or commented, pick out the people who match my ICP, then send a connection request and start a normal human conversation. No pitch, just a real chat. It works. It's just slow because I do all of it by hand. I want to automate the sourcing part. The workflow in my head: 1. Watch a list of creators. When one of them posts, catch it. 2. Pull everyone who engaged with that post (likers + commenters). 3. Enrich and score each person against my ICP (title, industry, company size, signals on their profile). 4. Every morning I get a CSV of the highest-scoring, most relevant people from the last day of engagement. 5. I do the human part myself: connection request + warm DM, zero automation on the actual messaging. The wall I keep hitting is LinkedIn itself. We all know how aggressive it is about automation, rate limits and restrictions. My account is literally my business, so a ban isn't "annoying," it's fatal. So my honest question for the people here who have actually built this kind of thing: If this were your system, how would you build it so it's safe and reliable? Specifically: - Where would you pull the engagers from? Apify actors, PhantomBuster, an unofficial API, a separate throwaway scraping account? What's actually low risk right now? - Would you keep the read-only scraping completely off my main profile, and only ever send connects from the main account? - What daily and weekly numbers do you treat as safe in 2026 for connects and DMs on a warmed account? - For scoring/enrichment, what are you using? Clay, a custom LLM step, something lighter? - And honestly, where's the line where this stops being worth automating and