Use two AI teammates to review one small project brief, then compare their responses in the same channel. This trial uses public sample text and asks for discussion only. It does not require shell access, file changes, Office, or Computer Use.
Follow Getting started to start the Web UI and daemon from source. In Settings → Model Workbench, configure your own chat model and run Check. Confirm a normal conversation returns a final response before adding teammates. Model checks and replies use your provider account and may incur charges.
Ling stores workspace state and history locally. The prompt and its context go to the providers you choose. This trial is not an offline-inference demonstration.
- Copy the sample project brief into a disposable local folder. Keep private code, credentials, and real customer data out of this first trial.
- In Settings → Teammate Management, configure two teammates. Give one a requirements-review role and the other an acceptance-review role. Bind each to a checked chat model; they can share a model for this trial, or use different models. A role or skill does not grant tools or file permission.
- Open Channels, create a project channel called Project review, select the disposable folder, choose a checked Ling Agent model, and add both teammates.
- Paste the brief into a message. Select a teammate through the channel's mention picker when asking for that teammate's review. Ask for discussion only; this first trial does not ask the agents to execute work or modify files.
Start with this prompt for the requirements reviewer:
Review the sample project brief below. Identify up to three missing requirements
that need a decision before implementation. Explain why each affects the first
version. Keep the scope small; do not implement anything or change files.
[Paste the Team Notes sample brief here.]
Wait for the requirements reviewer's final response. Then mention the acceptance reviewer through the picker in the same channel:
Review the sample brief and the requirements review above. Propose three concrete
acceptance checks for the first version. Separate required decisions from optional
improvements. Do not implement anything or change files.
Wait for the acceptance reviewer's final response. Finally ask the Ling Agent to summarize the two reviews, without mentioning either teammate:
Summarize the reviews in this channel as a short decision list: decisions needed
before implementation, three acceptance checks, and improvements that can wait.
Use the actual reviews; do not claim any code was changed or tested.
- Both requested teammates have a final response in the channel.
- The acceptance checks describe observable outcomes rather than general advice.
- The summary references the reviews and separates required decisions from later work.
- No file changes or execution are reported for this discussion-only trial.
This is an initial collaboration check, not a benchmark of model quality or an acceptance test for tool execution. Responses vary by provider and model. For later tasks that need tools, review tool bindings, workspace permissions, and approval requests first; then inspect actual artifacts and execution evidence.
Use the Issue forms and include your version or commit, OS, the first failing step, and a minimal reproduction. Share redacted errors only. Say whether you completed the review and which part of the result was useful; do not upload keys or workspace databases.
先按入门指南启动 Web 和 daemon，配置自己的聊天模型并通过 Check，确认普通对话能获得最终回复。模型检查和回复可能产生供应商费用。
把示例项目说明复制到一个临时目录。 在 Settings → Teammate Management 配置两个队友，分别负责需求边界和验收条件， 绑定检查通过的模型；然后在 Channels 创建项目频道并加入两个队友。
第一次只做讨论：粘贴示例文本，通过频道的成员提及选择器依次请两位队友评审， 等需求队友最终回复后再提及验收队友，等验收队友最终回复后，清除队友提及， 请 Ling Agent 汇总“实施前需要决定的事项、三个可观察的验收条件、可后置的改进”。 可以使用同一个模型，也可以分别配置模型；不需要开放 shell 或文件修改权限。
完成标准是两位队友均返回最终意见、汇总能对应实际评审，并且没有宣称修改或测试代码。 本地保存的是状态和历史，任务上下文仍会发送到你选定的供应商。把卡住的步骤和结果是否 有用反馈到 Issue 即可；源码 Alpha、讨论演示与原生工具验收是不同范围。