ParliAI

ParliAI is an AI aggregators tool. ParliAI was a multi-provider AI council that mixed models from Anthropic, OpenAI, Gemini and Grok into custom groups, then ran one-on-one answers, comparisons or multi-round debates with a full reasoning trail, scores and vote justifications.

ParliAI was a multi-provider AI council built for people who could not settle important questions with a single chatbot reply. The premise was simple: one model gives you an answer, several models give you a decision you can inspect. Users assembled a council from providers such as Anthropic, OpenAI, Gemini and Grok, then put a business or strategy question to the group and watched the models argue it out. The project is now archived and the service has closed, so what remains is a documented design and fictional sample conversations rather than a live product. The tool worked across three depth settings. One-on-one returned a single direct answer. Compare added answers from the other models in the council plus a curator pass that weighed them. Debate went further, repeating answers across rounds so each model could read the others and revise its position, at a much higher token cost. A curator model, optionally Grok, sat over the top of the council and produced scores, vote justifications and proposal diffs. Transparency was the central design claim. Every recommendation came with a full reasoning trail, so a user could follow how a conclusion formed, see which model pushed which argument and check the scoring rather than trusting an opaque verdict. The intended audience was decision makers, product teams and analysts who were already cross-checking prompts between tabs and paying for several subscriptions, and wanted that workflow consolidated under one plan. ParliAI sat in the same space as multi-model chat aggregators and AI debate or ensemble research tools, but leaned toward structured deliberation and auditability instead of fast single answers. With the service closed, it is best treated as a reference point for how multi-model consensus products were framed, not as a tool to evaluate for adoption. The examples on the archive are fictional and do not represent live output.