This page describes macOS and Windows desktop apps. See the guide for the WeChat Mini Program feature scope.
Two AI paths
YunStrat AI follows account access, quota and service status. The service manages its platform default model; ordinary users do not configure platform providers. Local / personal models retain Ollama, your own API credentials (BYOK) and configured Codex connections. Their setup and availability follow the relevant entry points and service terms.
The software is free to use. Third-party AI configuration is an optional paid feature; free features remain available without enabling it. Check the app for its activation process, scope and fee. The feature fee is separate from any model usage charges imposed by the provider.
Conversations and materials
The research agent supports multi-turn chat, paginated history, search, pinning and stop controls. Review the current destination, input, watchlists and enabled skills. Cloud chat, translation and polishing start from their explicit action buttons, without an additional request-preview dialog. Local / BYOK requests retain the interface preview and send confirmation. Local and cloud conversations are managed separately, without uploading all historical notes automatically. Personal models may use remote services: offline use depends on the selected recipient.
Cloud memory follows account preferences. A missing setting uses the current default; an explicit off preference is preserved. Watchlist and report access still need their respective business authorization. Enabling memory does not upload all local materials.
Draft polishing and saving results
Polishing processes the current conversation-input draft through the selected AI path. A complete, non-empty result is applied to the input automatically without sending; review the new draft before sending. Cancellation, failure or a changed original preserves the original text and does not send a message. Revisit research through conversation history and daily reports. The former Notes entry now opens AI daily reports; historical local research records can still be imported and exported in Settings.
Preview skills before installation
Local and cloud skills are saved separately. Command installation downloads skill text for source and content review, followed by confirmation. New skills are disabled by default. It does not run shell, npm or skill scripts or grant extra trading, tool or file permissions.
Where skills are supported, use @ in the conversation assistant to select enabled skills for the current destination. Local and cloud candidates are read separately; the cloud does not automatically read or upload local skills. Cloud temporary conversations do not use research skills. A referenced skill is research-method material, not permission to execute commands or access additional tools.
Daily plans and task states
Daily reports require account access and valid plans, with notifications using the configured channels. Review history and running, completed or failed states. Eligible failed tasks support plan verification and confirmed retry, which may consume quota. Acceptance is not completion; stopping observation or locking the app does not cancel an accepted cloud task. Completed results distinguish server-AI reports from basic data briefs generated without AI, and show actual coverage. Completion does not establish notification delivery.