/
Work stream 1: Geo-distributed Cloud for AI agents

Work stream 1: Geo-distributed Cloud for AI agents

  • Leaders: @Yona Cao @C.C. Fan @Felix Sun

  • Objective: This work stream focuses on optimizing the performance and scalability of cloud infrastructure across multiple geographical locations. It aims to address challenges related to latency, data sovereignty, and efficiency in distributed computing environments.

  • Approach: The team will develop strategies for seamless data synchronization and application performance across dispersed networks, ensuring robust, secure, and compliant operations globally.

 

Workstream 1 Master Overview: Geo-Distributed Cloud for AI Agents

Author: Tina Tsou

Workstream Leaders: @Yona Cao, @C.C. Fan, @Felix Sun

1. Executive Summary

Objective

Optimizing the performance, security, and scalability of cloud infrastructure across multiple geographical locations for AI agents. This workstream addresses latency, data sovereignty, and compute efficiency in highly distributed environments.

Approach

  • Milestone 1 (Local Foundation): Developing a secure, isolated local multi-agent runtime environment at the edge to handle immediate tool-calling and context constraints.

  • Milestone 2 (Global Distribution): Utilizing vivgrid, a geo-distributed public edge cloud, to deploy and accelerate AI agent applications globally, bringing inference and serverless execution closer to users.

image-20260520-062521.png

 

2. Milestone 1: Multi-Agent Local Framework

Purpose

Before agents can run across high-latency global networks, they must execute safely and efficiently at the local edge. Milestone 1 establishes the local runtime node environment that guarantees security and deterministic latencies.

Core Concepts

  • Supervisor-Worker Pattern: A primary "Orchestrator" breaks tasks down and delegates them to specialized, isolated "Sub-Agents."

  • Context Isolation: Sub-agents execute tasks in separate sessions and return only distilled, high-signal results. This keeps the primary model's context window clean, reducing hallucinations and API costs.

image-20260520-062543.png

 

Deliverables

  1. Academic Paper: The Ontological Executive Architecture: Bounding Local Contexts to Subvert Self-Attention Complexity in Edge-Native Systems

  2. Framework Implementation: SuperAI Local Edge-Agent Runtime (SLEAR) (A functional runtime with a Local Orchestrator Engine, Model Context Protocol integration, and sandboxed WebAssembly tool execution).

3. Milestone 2: Geo-Distributed Public Edge Cloud (vivgrid)

Milestone Leaders: @Yona Cao, @C.C. Fan

Purpose

vivgrid.com is a geo-distributed public edge cloud built to run AI applications with global AI inference infrastructure, deploying agent apps closer to users for low-latency performance.

image-20260520-062559.png

 

Key Features

  1. Global AI Inference Infrastructure: 1-click application deployment across 25 cities in 5 major regions, routing users automatically to the nearest compute node for sub-200ms Time-to-First-Token (TTFT).

  2. AI API Bridge: Accelerated local entry points and routing for US-based LLM APIs (like OpenAI), allowing developers to test and call multiple APIs seamlessly in a single codebase.

  3. Function Calling Platform: Serverless environment where developers easily configure tool-calling once. Vivgrid translates and handles compatibility between different LLM backends.

  4. Prompt Management: A simple dashboard allowing non-developers to instantly update system prompts globally without changing code.

Deployment Timeline

  • Phase 1 (API Acceleration): Launch the OpenAI API acceleration gateway across all 25 edge nodes to optimize streaming requests and responses.

  • Phase 2 (Serverless Tool Integration): Roll out the serverless function-calling engine, enabling agents to call third-party APIs (e.g., weather APIs) and access private data with high-speed local performance.