Featured Project

EOS AI

Purpose-Built Assistant for EOS Businesses

An intelligent assistant trained on core EOS materials to help businesses stay on track between implementer sessions.

Next.js 15 Vercel AI SDK Claude 4.5 Sonnet / Opus 52 Tables
Explore the project

Project Overview

Built  for  the  leaders  who  run  the  world's  best  businesses

EOS Worldwide provides the Entrepreneurial Operating System, a complete set of simple concepts and practical tools used by thousands of companies worldwide to clarify, simplify, and achieve their vision.

EOS AI is an intelligent assistant trained on core EOS materials and best practices. It helps businesses stay on track with their EOS implementation by providing instant answers, guided tools, and session preparation support.

Important: EOS AI is designed to complement EOS Implementers, not replace them. Your implementer remains essential for strategy, accountability, and the human element of EOS. EOS AI simply helps you apply what you've learned between sessions.

52 Database Tables
14 Feature Modules
6 Knowledge Sources
9 Tool Types

Visual Showcase

See  the  platform  in  action

Technical Architecture

Under  the  hood

Assistant Architecture

  • Dynamic Model Selection via preflight analysis (Claude 4.5 Sonnet or Claude 4.5 Opus)
  • Complexity Detection adjusts reasoning effort (low/medium/high)
  • 9 Tool Types including documents, search, and calendars
  • Persistent Memory across conversations

6 Knowledge Sources

  • System Knowledge - Core EOS materials (Priority 1)
  • User Memories - Explicit facts to remember (Priority 2)
  • Persona Documents - Role-specific knowledge (Priority 3)
  • Conversation Summary - Long-context continuity (Priority 4)
  • User Documents - Uploaded files & data (Priority 5)
  • Company Context - Business metadata (Priority 6)

Database

  • 52 PostgreSQL tables via Drizzle ORM
  • Users, Organizations, Chats, Messages
  • Personas, Memory, Research Sessions
  • L10 Meetings, Voice Recordings

Real-time

  • Resumable Streams via Redis persistence
  • Context Budgeting with adaptive token limits
  • Version History with undo/redo
  • Live Research progress in Nexus mode

Technology Stack

Built  with  modern,  scalable  technology

Frontend

Next.js 15.3 App Router + RSC
React 19 RC Server Components
shadcn/ui Radix Primitives
ReactFlow Org Charts
GSAP + Framer Animations

Assistant / ML

Vercel AI SDK 4.3 Streaming + Tools
Claude 4.5 Sonnet Primary Model
Claude 4.5 Opus Complex Queries
Upstash Vector 1536-dim Embeddings
pgvector HNSW Indexes
Whisper API Transcription

Backend

PostgreSQL Neon / Vercel
Drizzle ORM Type-safe Queries
Redis Resumable Streams
Zod Schema Validation

Services

Auth.js 5.0 OAuth + Credentials
Stripe Subscriptions
Vercel Blob File Storage
Resend Transactional Email

Intelligent Retrieval

RAG  Made  Easy

All knowledge sources queried in parallel during preflight for the best possible answer

? Your Question
System Knowledge Core EOS materials
User Memories Facts you've shared
Persona Context Role-specific knowledge
Conversation Current context
Your Documents Uploaded files
Web Search Real-time results
Parallel Queries
Priority-Ranked
Source Tracking
Auto Deduplication

Preflight Engine

Parallel  Intelligence,  Zero  Lag

Every request is analyzed in parallel before a single token is generated. The preflight engine decides model configuration, token budgets, and tool strategy in under 200ms.

Incoming Query Validated and parsed
6 Parallel Retrievals All knowledge sources queried simultaneously
System Memory Persona User Docs Org Company
Preflight Decision Claude 3.5 Haiku in <256 tokens
Thinking Budget Token Limit Tool Strategy
Optimized Stream Model + tools pre-configured

Lightweight Model Pass

Claude 3.5 Haiku analyzes query complexity with a single 256-token call. Deterministic temperature zero output -- no wasted generation cycles.

Adaptive Token Budgets

Dynamically allocates 1K-64K output tokens per request. Extended thinking budgets scale from 16K to 64K based on detected complexity.

Early Tool Steering

When document creation is predicted, the first model step is force-directed to the correct tool -- eliminating unnecessary reasoning loops.

Developer API

OpenAI-Compatible  REST  API

Full-featured API surface under /api/v1 with per-key rate limits, SSE streaming, and scoped access control.

Bearer Token Authorization: Bearer eosai_sk_...
API Key Header X-API-Key: eosai_sk_...
Rate Limits 60 RPM / 1,000 RPD per key
POST
/chat Chat completions with EOS context
SSE
POST
/embeddings 1536-dim vector embeddings
ada-002
GET
/models Available model tiers
v1 fast pro
GET
/usage Rate limits and consumption stats
Headers
POST
/conversations Persistent multi-turn sessions
SSE CRUD
POST
/documents/analyze Document Q&A with chunked retrieval
SSE Multipart
eosai-v1 Default -- best balance of speed and quality
eosai-v1-fast Optimized for quick, lightweight responses
eosai-v1-pro Enhanced reasoning for complex scenarios

System Architecture

End-to-End  Request  Lifecycle

From user input to persisted response -- every request flows through a layered pipeline of validation, retrieval, generation, and persistence.

User Input
Auth + Validation
Chat Route
Parallel Fan-Out
RAG Retrieval 6 sources
Preflight Haiku
Tool Wiring 9 tools
Mentions Smart detect
Generation
Claude 4.5 Sonnet streamText / Nexus
SSE Stream Redis buffered
Persistence
PostgreSQL 52 tables
Redis Stream buffer
Upstash Vector 1536-dim
Vercel Blob File storage

Interested  in  building  something  similar?

Let's discuss how I can help bring your enterprise product vision to life.

Get in Touch Back to Portfolio