From passcode to fused review in about five minutes.
Open /login and enter the team passcode. If your deployment has no admin token configured, you'll be let straight through — public deployments always should set one.
In Settings, add at least one OpenAI-compatible provider — base URL, model name, and API key. With no keys configured the app runs in demo mode with clearly labeled sample output.
On the main page, type your topic (at least 2 characters), pick a review mode, and choose how many models should participate — the full pool or a random subset.
Hit 开始并行搜索. Each model answers in its own column, streaming in real time. You can stop early with 停止搜索 — partial results remain usable.
When all models finish, click 查看模型对比表. The comparison matrix shows which scholars, works, and claims appear across models (consensus) versus uniquely.
Click AI 融合总结 and an adjudicator model weighs all outputs into a single synthesized review. Export it as Markdown or PDF from the fused page.
An 80-second walkthrough of a real run: choosing the 文学启发 output format, picking 3 models, streaming results in parallel, generating the AI 融合总结 essay, and exporting the watermarked PDF — all as an anonymous guest, no sign-in needed.
Want to see the exported file? Download the sample PDF — guest-tier exports are image-based and watermarked; signed-in Pro and above get text-based PDFs.