BidBand: The RFP War Room

Answering an enterprise or government RFP is the most painful cross-team workflow in B2B sales: 20–40 staff-hours per bid across sales, experts, writers, compliance and an approver, and roughly half of all disqualifications come not from weak answers but from a single missed mandatory requirement. BidBand drops an RFP into a Band room where five specialised agents work the bid in real time. Maestro parses the RFP and publishes the mandatory-requirement matrix. Scout mines the company knowledge base for proof points. Quill drafts each section and hands it peer-to-peer straight to Gavel. Gavel audits every draft against the matrix and can veto non-compliant work, bouncing it back with findings while you watch the rejection-to-revision loop happen live. Sentinel assembles the proposal, scores 100% requirement coverage, seals a hash-chained audit log, and asks the human bid manager to sign off. Band is the only channel the agents use: every handoff, recruitment and tool call flows through the room as @mention-routed messages and events. The transcript IS the workflow. The originality hook is cross-provider checks and balances: the writers (Maestro, Quill) run on AI/ML API closed-weight models, while the researcher and reviewer (Scout, Gavel) run on Featherless open-weight models. No provider ever approves its own work. The demo voiceover is generated with ElevenLabs. Pure Python standard library, zero pip installs: judges run the entire five-agent flow — including a guaranteed compliance veto — with no API keys (python run.py).