Prototype Deployed in Production
IVE Strategic Execution Agent
The reference implementation of the IVE engine, built for the All Things Agentic Hackathon 2026. It receives a strategic goal, loads real context through Supabase, reasons with Gemini, creates an action, verifies persistence with a read-after-write check, and records the decision in business memory.
From goal to verified action
Technology
Model
Gemini 3.5 Flash
Agent orchestration
Google ADK (LlmAgent, Runner, RunConfig)
HTTP service
FastAPI — GET /health, POST /run
Hosting
Google Cloud Run
Authentication
Supabase JWT, validated via JWKS (ES256)
Data
Supabase, protected by Row Level Security
Demonstration video
What this is not
The implementation does not claim automatic retries, exactly-once guarantees across different requests, persistent ADK sessions, learning over time, or user rate limiting.
The write to business memory is correct by code review but has not yet been exercised by a dedicated production smoke test.
InsightValues is a pre-existing Flutter and Supabase platform. This IVE Strategic Execution Agent is the new reference component built specifically during the All Things Agentic Hackathon 2026.
The identity and Row Level Security model described above is analyzed in depth, as a general pattern, in AI Agents Don’t Need More Permissions. They Need Row-Level Security.
The write_agent_memory and get_project_context mechanism described above is examined as a general pattern — including its honest limitations — in AI Agent Memory in Production.
The orchestration layer itself (Google ADK) has moved on since this was built — see what changed in ADK 2.0 and what that would mean for a Human Gate built today.