Official Accounts
- Official X: @aiwolfdial_nlp
Announcements about the contests and updates to this site. - Official GitHub: github.com/aiwolfdial
The organization that hosts the repositories of the Natural Language Division. - AIWolf Project: aiwolf.org
The official website of the AIWolf project as a whole.
Log Viewer (aiwolf-nlp-viewer)
A viewer for watching and replaying game logs in the browser. No installation is needed; use the hosted version.
- Hosted version: aiwolfdial.github.io/aiwolf-nlp-viewer
- Source code: aiwolf-nlp-viewer
How to use it:
- Open a published log directly: open a “Log list” from Game Logs by Track (links for every contest and track) or from a contest’s results page, and press “▶ 開く” (open) in the row of the log. The viewer opens with that log loaded.
- Choose a bundled log: on the viewer’s archive page, choose Select Folder (contest) and then Select Log (file) to view the logs used for the human evaluation of past contests.
- Open a local file: on the same archive page, select a file, or copy the text of a log and press “Paste from Clipboard”. Logs from your own server can be viewed this way too.
- Share by URL: a link of the form
https://aiwolfdial.github.io/aiwolf-nlp-viewer/archive?url=<URL of the log>opens that log (published log server only).
Official Repositories
Repositories used for agent development. Please refer to each README for setup and usage.
| Repository | Description |
|---|---|
| aiwolf-nlp-agent | Sample agent |
| aiwolf-nlp-agent-llm | Sample agent using an LLM |
| aiwolf-nlp-common | Common package for agents |
| aiwolf-nlp-server | Game server |
| aiwolf-nlp-viewer | Log viewer (see above) |
| aiwolf-nlp-llm-judge | LLM-as-a-Judge relative evaluation of game logs |
How to Browse Past Results and Logs
Results and logs of the contests since 2024 are collected on Results & Logs.
- Pick a contest: open a contest card on the Results & Logs page to reach its results page. The same page is also reachable from “Results & Logs” in each contest’s menu.
- Page structure: official results (awards), win rates, game metrics, human evaluation, LLM-as-a-Judge (relative evaluation) and game logs, with a table of contents at the top. Each section starts with a short explanation of the metric.
- Across contests: the “Across contests” group at the top of the Results & Logs page has four pages: awards by contest, evaluation methods (judges and models), agreement with human evaluation, and game logs by track.
- View or save logs: Game Logs by Track or the “Game Logs” section at the end of each results page links to the log listings on the public log server. In a listing, “▶ 開く” opens a log in the viewer, “⬇ 保存” saves one file, and “⬇ zip” in a folder row downloads the whole folder as a zip.