The Problem
Market-moving events are ambiguous. A Fed rate hike is bullish for banks, bearish for growth stocks, and neutral for commodities — depending on who you ask. No single analyst captures all of that. The interesting signal isn't any one view; it's where a diverse set of participants land after they've heard each other out.
Approach
Swarm runs 15 distinct trader personas in parallel — each with a different role, style, natural bias, and time horizon. A quant fund doesn't react the same way as a macro tourist or a retail momentum trader.
Round 1 is independent: every agent reads the headline and forms a view without knowing what anyone else thinks. Round 2 is the debate: each agent reads a summary of all other positions and decides whether to hold firm, partially update, or flip entirely. View changes are tracked and surfaced in the final output.
The result is an emergent consensus — or a persistent disagreement — that no single prompt could produce.
Why Two Rounds
A single round of parallel calls gives you a breadth of views but no interaction. The interesting dynamics in real markets happen when participants update on each other. A momentum trader who reads that every macro fund is bearish might temper their enthusiasm. A contrarian who sees consensus forming in one direction has more reason to go the other way.
Round 2 is what makes the output feel alive rather than just a list of opinions.
Design
The entire tool is a single Go binary with no runtime dependencies. Agents run concurrently via goroutines. The output is a color-coded terminal display with a signal bar chart, sentiment breakdown, and a watch list of the most-mentioned assets.
Install via the releases page or go install — no config files, no accounts, just an API key.