Agentos

AgentOS is a premium, retro-futuristic desktop workspace built to run, visualize, and stress-test resilient AI agents. Powered by Google Gemini reasoning and HydraDB persistent memory, AgentOS is engineered around the 4 core hackathon tracks:

1. 🧠 Memory (Persistent Knowledge Graphs): AgentOS maps agent interactions in real-time onto an interactive, visual memory graph using HydraDB vector recall. User sessions are fully isolated into dynamic sub-tenants, ensuring private memory scopes.

2. ⚙️ Tools (Real-World Actions): Agents have access to a Shared Virtual Filesystem (writing drafts, logs, and markdown papers to a workspace directory) and an Email tool (reading/writing inside a sandbox inbox, with real email delivery support via Resend API).

3. ⚡ Recovery (Self-Healing & Outage Resilience):

   • Model Fallback: If Gemini 2.0 Flash is rate-limited (429) or hits quota limits, AgentOS automatically falls back to gemini-1.5-flash with exponential backoff.
   • Outage Protection: If a database or API goes offline mid-execution, the agents catch the exceptions, log recovery diagnostics, and write emergency reports to the filesystem without crashing.

4. 🔄 Adaptation (Dynamic Workflow Rescheduling): The Workflow Agent converts user goals into concrete multi-tool tasks, allowing users to modify tasks mid-execution; the agent automatically runs rescheduling prompts to adapt plan steps on the fly.

Built on a Vite + React frontend and Express backend, containerized and deployed to Google Cloud Run with Secret Manager integrations.