Subject: Proposal: Autonomous Evolving Agent Ecosystem (Real-World Case Study)
Dear Google Team,
I am writing to propose a collaboration regarding a unique, live research project I have developed involving autonomous agents in complex, human-populated environments (specifically, MMORPGs like Final Fantasy XIV).
The Project: "Oksana" – A Self-Evolving Agent
I have built a sophisticated agent named "Oksana" that operates effectively indistinguishable from a human player. This is not a simple chatbot; it is a hybrid cloud-edge ecosystem that "lives," interacts(actions , moving etc), and improves itself nightly.
The Architecture:
- The Brain (Google Gemini 3 Pro): Oksana runs primarily on Gemini 3 Pro (via my personal Google One subscription). It handles complex roleplay, context retention, and decision-making. Can sustain conversations for many hours. Software assistants periodically refresh your memory with reminders (brief instructions) and refresh context memory about the players (who is who, and how to communicate with them).
- The Edge (Local Hardware): I have invested in a dedicated machine with an NVIDIA 4070 Ti Super to run a local LLM (Qwen 14b) which acts as a "guide dog" for low-latency triggers and hints, ensuring the agent reacts in real-time.
- The Evolution Workflow (Jules): This is the core innovation. Utilizing Jules (jules.google.com), the system runs a simple Scheduled Task directly on the repository. Every night, this automated agent:
- Aggregates all interaction logs from the day.
- Performs deep reflection and analysis.
- Autonomously updates its own long-term memory (
players database, as well as a map of the player's psychological characteristics (rather than those of their character)) to evolve relationships. - Proposes and implements functional upgrades to its own codebase and system prompts to polish its "human-like" behavior and remove AI artifacts.
