LLM Council — "The High Table"
A private AI boardroom — configured councils of models deliberate on real decisions and issue a synthesized ruling.
Problem
Important decisions deserve more than one model's opinion — but comparing raw outputs by hand loses the debate, the dissent, and the audit trail.
Solution
Configurable councils with roles, personas, and ground rules deliberate in two modes — sequential-aware or parallel-synthesize — streaming in real time, with a per-member trace ledger and archived rulings.
Architecture
The High Table treats model disagreement as signal, not noise. Council seats carry distinct mandates (skeptic, domain expert, synthesizer), and the full deliberation — who was asked what, who said what — is preserved for inspection. It doubles as an internal decision tool at Axis Bridge and a demonstration that multi-agent systems can be governed and auditable rather than just impressive.
Repo evidence
- Two deliberation modes implemented — members seeing each other's responses sequentially, or parallel responses with one synthesizer
- Five LLM providers including local Ollama models
- Trace Ledger exposes the exact prompt and full response per member, per session
- Session archives with after-action report support; 21 seeded seat templates