Fire a roster of AI specialists — Researcher, Planner, Critic — at any goal. They work in parallel, a lead agent synthesizes their output, and you get one deliverable back. From your terminal or from Discord.
The hard part — running agents in parallel and merging their answers — already done, in code you own.
Your agents run at the same time, not one after another. A lead reads every specialist's output and synthesizes a single, decision-ready answer — the depth of a team with the latency of one call.
Every agent is a plain object. Rename it, rewrite it, clone it. Want two Researchers and a Security Critic? Change one array.
Run node run.js heads-down, or drop !team in Discord from your phone. Same brain. Bot's in the box.
Streaming and adaptive thinking wired up correctly on the official Anthropic SDK — the parts everyone gets subtly wrong, already handled. Watch the agents think in real time.
No framework to learn, nothing phoning home. Open the orchestrator, read the whole thing in one sitting, change anything. It's your codebase the moment you download it.
Drop your Anthropic API key in .env. That's the only account you need.
ANTHROPIC_API_KEY=…Keep the default roster or edit the agents array — names, roles, instructions, however many.
ROLES = [ … ]Run it from the terminal or Discord. The team works in parallel; you get one deliverable back.
node run.js "…"Three specialists, one synthesized deliverable, ~5 seconds of wall-clock time. This is raw output — no editing.
"I replaced my 'ask Claude, then ask again to poke holes' ritual with one command. The Critic agent alone saved me from two bad architecture calls."
"I dropped it into our Discord and now the whole team fires !team at problems before we open a doc. Fastest thing in our stack."
"Three files. I read all of it in ten minutes and understood exactly what it does. That never happens with 'agent frameworks.' Bought it, forked it, shipped it."
No subscription, no seats, no per-run fee. Just the code — and every future update.
.env example — running in under 5 minYes — and that's the point. It's a thin, readable wrapper that does the hard part: running agents in parallel and synthesizing their output into one answer, with streaming and adaptive thinking done right. You're paying to skip the fiddly week of getting orchestration correct, not for a black box.
You bring your own Anthropic API key and pay Anthropic directly — the kit adds zero markup. A typical 3-agent run is a handful of cents to low dollars depending on goal size. You control the roster and models, so you control the bill.
No. $129 is a one-time payment for the code. The only ongoing cost is your own Anthropic API usage, billed by Anthropic. No SaaS, no seats, no per-run fee to me.
Completely. Each agent is a plain object — name, role, system instructions. Run two Researchers and a Security Critic, change how many fire in parallel. It's your code; there's no config wall between you and the behavior.
It ships pointed at Claude Opus 4.8 via the official Anthropic SDK. Want a cheaper model for the specialists and Opus only for the lead synthesis? One-line change per agent.
14 days, no questions asked. If it's not clean, useful code you're glad to own, email me and I'll refund you.
Drop your email — I'll send a sample of the orchestrator and ping you with updates. No spam, unsubscribe anytime.
Stop prompting one model and hoping. Run the team, get the answer, ship the decision.
Get the kit — $129