Armadra 会替我调用模型吗？

不会。Armadra 启动你本机已经安装并登录的 CLI，模型调用由各 CLI 按其提供方的规则完成，费用也按你在各提供方的账户计算。协调者 ama 的模型密钥保存在本机的密钥后端，只交给 ama 进程。

我的数据存在哪里？

桌面安装没有服务端。工作空间记录保存在本机 SQLite 与工作区的 .armadra/ 目录。凭据与终端原始输出不会写进画布、日志或接口响应。

会改动我的 CLI 配置吗？

不会。Hook、技能和画布说明只在从画布启动 CLI 时注入，画布外启动的同一个 CLI 不受影响，Armadra 也不写 CLI 的全局配置。

关掉窗口后 Agent 还在跑吗？

在。关闭窗口不停止后台；使用 tmux 后端时，终端会话还能跨应用重启存活。

数据库损坏怎么办？

Armadra 遇到未知或损坏的数据库会拒绝启动，不会自动清空或重建，以免丢数据。可以用设置里的一致性备份恢复。

为什么系统提示安装包有风险？

当前安装包还没有代码签名与公证，见安装里的处理方法。签名发布在路线图上。

和 Workflow 工具有什么不同？

Workflow 擅长把步骤明确的任务编排成可重复流程；Armadra 的节点运行的是完整的 Agent 产品，适合路径会随发现变化的开放任务。两者可以结合：先在画布上探索，再把稳定的协作存成工作流。

在哪里反馈问题？

软件问题请到 AMA-Link/Armadra；官网与文档的问题请到 AMA-Link/armadra-site。

Does Armadra call models for me?

No. Armadra launches CLIs that are installed and signed in on your machine; each CLI calls models under its provider’s rules, billed to your own provider accounts. The ama coordinator’s model keys stay in the local secret store and are handed only to the ama process.

Where is my data?

The desktop install has no server. Workspace records live in local SQLite and the workspace’s .armadra/ folder. Credentials and raw terminal output never go into the canvas, logs or API responses.

Will it change my CLI configuration?

No. Hooks, skills and canvas notes are injected only when a CLI is launched from the canvas; the same CLI started outside is unaffected, and Armadra doesn’t write the CLI’s global configuration.

Do agents keep running after I close the window?

Yes. Closing the window doesn’t stop the background; with the tmux backend, terminal sessions also survive app restarts.

What if the database is damaged?

Armadra refuses to start on an unknown or damaged database instead of wiping or rebuilding it, so nothing is lost silently. Restore from a consistent backup made in Settings.

Why does my OS warn about the installer?

Packages are not code-signed or notarized yet; see Install for how to proceed. Signed releases are on the roadmap.

How is it different from workflow tools?

Workflow tools excel at turning well-defined steps into repeatable pipelines. Armadra’s nodes run complete agent products, which suits open-ended tasks whose path changes as you learn. They combine: explore on the canvas first, then save the settled collaboration as a workflow.

Where do I report problems?

App issues go to AMA-Link/Armadra; website and docs issues go to AMA-Link/armadra-site.