Band of Agents Hackathon 2026

VERDICT
ROOM

An AI courtroom for buying decisions. Six agents debate a vendor choice inside a Band room — and return a scored, defensible verdict in minutes.

BUY
BUY WITH CONDITIONS
AVOID
The problem

Vendor due diligence is broken.

Choosing a B2B tool means one person juggling pricing, security, GDPR and migration — then writing a single biased summary with no real review.

20–40
hours
of manual research per vendor
2–4
weeks
spread out, blocking the decision
1
biased summary
no adversarial check, no paper trail
How it works

Six agents. One Band room. A scored verdict.

User drops
a case
BAND ROOM
ResearcherScout AdvocateCritic ComplianceArbiter
VERDICT
Score + Conditions + Dissent

A Researcher gathers sourced facts, a Scout finds alternatives, an Advocate and a Critic debate two rounds, and an Arbiter issues the verdict — like an investment-committee review, but in minutes.

What's new

Two things no one else is doing.

1

Dynamic recruitment

When the Critic flags a compliance risk, the Arbiter pulls a Compliance specialist into the live Band room mid-debate. Agents recruiting agents — not pre-wired.

2

Cross-model courtroom

Debaters run across DeepSeek-V3.1-Terminus, DeepSeek-V3.2 and GPT-4.1-mini over AI/ML API and Featherless — a genuine multi-model panel, not one model talking to itself.

Live example

Notion vs Confluence vs Slite.

A 50-person EU software company asks: adopt Notion as the main workspace? Priorities: cost, migration, security, GDPR. The GDPR concern triggered a live Compliance recruitment.

75/100
BUY_WITH_CONDITIONS
Value for money
18/25
Capability fit
22/25
Risk profile
20/25
Alternatives
15/25
Read the full verdict →
Built on

Band is the coordination layer.

Every handoff, shared fact, debate turn, recruitment and verdict travels through one Band room — not a wrapper, the actual collaboration layer.

Band SDKAI/ML APIFeatherless DeepSeek-V3GPT-4.1-miniLangGraphPython