Heavy Rain
Heavy Rain turns recurring work across AI assistants into shared internal applications.
Core trick
Employee-approved chat excerpts are clustered for repeated needs, then a sandboxed agent builds candidate apps on shared components after human release review.
Description
Heavy Rain turns recurring work across AI assistants into shared internal applications. From employee-approved weekly AI chat excerpts, the system identifies repeated needs, checks existing company apps first, and either reuses a matching tool, requests missing evidence, or commissions a sandbox app when a gap exists. Specialized agents build each candidate on shared company UI components and technology, security best practices, and SSO... pairing verification checks with human release review so teams get tailored tools that remain consistent across their organization.
dominicphillips/argon-hackathon-heavy-rain★ 10 forksTypeScriptCreated Oct 25, 2026 · last push Oct 26, 2026
README excerpt
# Heavy Rain Heavy Rain is an internal tool platform for teams working across AI assistants. Employees submit explicitly reviewed AI workflow reports through **Relay** (intake and builder coordination). When a build is commissioned, **Relay** can launch a live sandbox build through the Google Gen AI Python SDK (`antigravity-preview-09-2026`, `environment="remote"`)—requesting either a quick `Flash` wireframe scaffold (`gemini-3.8-flash`) or an `Argon` high-fidelity interactive application (`gemini-4-argon`)—and serve the resulting self-contained static HTML/CSS/JS bundle locally for human review. **Talos** (**work in progress**, replacing the earlier `Cairn` product name) is Heavy Rain's planned workflow discovery and decision system for grouping repeated weekly work, checking the approved application catalog first, and routing each recurring need to reuse an existing catalog app, request missing schema or scope evidence, commission a sandbox build, defer to the next cycle, or decline a build. In the current repository, weekly routing runs deterministic fixture rules in `services/cairn` (retained as the Python package and API identifier for compatibility). Every workflow follows the traceability chain `approved snippets -> weekly work -> evidence and counterevidence -> catalog fit -> decision -> tooling -> verification` ([`__grounding/demo/weekly-work-to-apps.md`](./__grounding/demo/weekly-work-to-apps.md)). All people, avatars, verbatim quotes, organizations (**AUGI** and