AURORA · AUTO ROUTER / LOW AI
One conversation. Review the model and token range before running.
Aurora (Auto Router · Low AI) is a desktop AI workspace for reviewing request scope, execution models, and estimated input and output tokens. The browser demo is a local preview without external AI calls; the interactive interface is currently in Korean.
What you can explore
One conversation with organized work
Keep work in a continuous conversation, group related messages, archive or move work to trash, and restore it later. A legacy mode also supports manually managed sessions.
Compare models before execution
Configure separate analyzer and executor roles and choose the model for each request. In the public developer preview, analysis and model selection are manual; automated routing quality is not guaranteed.
Separate estimates from observed usage
Review input and output token ranges based on character counts and tier-level output assumptions. Compare with a planning budget you set; these figures are not account quotas or guaranteed costs.
macOS and Windows installers
Download installers for Apple Silicon and Intel Macs and Windows x64. Windows can download new versions automatically. macOS updating requires a Developer ID-signed build.
Frequently asked questions
Does the browser demo call an AI provider?
No. It previews conversation organization, model comparison and estimated token ranges locally. Provider API connections are configured in the installed desktop app.
Does Aurora automatically switch execution models?
The public 0.1.2 preview requires an explicit analysis action and manual execution-model selection. A recommendation is separate from an execution request.
Are token estimates exact costs?
No. They use tier-level output assumptions, not per-model empirical predictions or cost guarantees. Actual usage returned by a provider is displayed separately.
Where is conversation data stored?
Conversation persistence is off by default. Enabling it stores data on the current device; it is not an encrypted vault. The web preview does not send conversations to AI providers.
How do updates and installation warnings work?
Windows updating is included from 0.1.2. Versions 0.1.1 and earlier need a manual replacement. The current ad-hoc macOS preview has automatic updates disabled, and operating-system security prompts may appear.
Developer preview limitations
Only text responses are supported; filesystem editing and shell execution are unavailable. Whole-task token and cost enforcement and real cross-version updating remain unverified. Task completion and search inclusion are not guaranteed.
Installers · Changelog · MIT license