Memory That Preserves Evidence and Context
Separates and preserves users’ work history, decisions, permissions, and state in a memory layer, restoring them with sources when needed.
GONGJAK RESEARCH / APPLIED AI
Not AI built from scratch,
but AI that transforms systems already at work.
GONGJAK explores how to transform existing services, business tools, and personal workspaces into agent interfaces, built on a shared foundation of long-term memory and AEO.
RESEARCH INDEX04
COMMON CORELONG-TERM MEMORYAEOCONTROLLED AGENTS
COMMON RESEARCH CORE
Four research directions.
One design principle.
Long-term memory and AEO form the knowledge base. Controlled execution and integration with existing systems form the shared technical approach.
Knowledge Built for People and AI
Structures public information for clear answers, enabling website agents and external answer engines to use the same verified sources.
Controlled Agent Execution
Uses strict scope limits, permission checks, independent watchdogs, and audit logs to keep agents within authorized information and actions.
Connect Before You Replace
Adds adapters and an agent layer to existing interfaces, APIs, and data to reduce rebuild costs and migration risks.
RESEARCH DIRECTIONS
These are current research, prototype, and concept directions. Scope and specifications may change during validation.
AX Bridge: Turning Existing IT Tools into Agents
Rather than rebuilding the IT tools already used by governments and businesses, we are researching an integration layer that connects them to proven LLMs and major platforms. The goal is to reduce AX transformation costs by wrapping existing interfaces, APIs, and data as agent tools, with added permissions, approvals, and auditing.
- Adapters for existing interfaces, APIs, and data
- Designs that convert workflows into agent tools
- Management tools with permissions, approvals, and audit logs
- Minimal rebuild scope and AX transformation costs
INTEGRATION MAPLIVE MODEL / RESEARCH
01EXISTING SYSTEM
02AGENT LAYER
03PROVEN AI TOOLS
EXPECTED DIRECTIONLOW-COST AX
A Company Information Agent for Existing Websites
Helps visitors explore company, service, and project information already available on a website using natural language. It can be added without rebuilding the site, using only an embed script and APIs, and connects to an AEO knowledge structure so people and AI search systems understand the same sources.
- Cost-efficient LLM routing and verified knowledge retrieval
- Thresholds that block out-of-scope questions
- An independent watchdog that detects threshold bypass attempts
- Existing website ingestion and updates linked to an AEO structure
SAFE ANSWER FLOWLIVE MODEL / PROTOTYPE
01SITE CONTENT
02THRESHOLD + WATCHDOG
03VISITOR ANSWER
EXPECTED DIRECTIONNO REBUILD
BLE- and GPS-Based Multicamera Sports Highlights
Combines footage from multiple cameras, venue BLE reference points, and player GPS data into a single timeline. We are exploring a recreational sports experience that analyzes position and movement during play, then automatically edits footage from four angles for personal download and sharing after the game.
- Automatic synchronization across N cameras
- Venue scale calibration using BLE reference points
- Position and movement analysis based on player GPS
- Automatic four-angle highlights and mobile sharing
MULTI-SENSOR TIMELINELIVE MODEL / CONCEPT
014× CAMERA
02BLE + PLAYER GPS
03AUTO EDIT
EXPECTED DIRECTIONMOBILE HIGHLIGHT
A Personal AI Server for Owning Your Long-Term Memory
Preserves a personal workspace and long-term memory on a compact Raspberry Pi-class server, connecting to commercial LLMs for reasoning when needed. This research combines Docker-based apps with a secure remote access path to create a compact, portable, and recoverable personal AI workspace.
- User-owned long-term memory storage
- Clear separation of roles between commercial LLMs and local data
- Docker-based personal workspaces and service hosting
- Secure remote access through Cloudflare Tunnel
MEMORY × AEO
Remember on the inside.
Answer accurately on the outside.
PRIVATELong-term memoryWork history · Decisions · State · Permissions
→CONTROLAgent LayerRetrieval · Threshold · Watchdog · Action
→PUBLICAEO knowledgeAnswer content · Evidence · Structured data
Personal memory stays inside its permission boundary, while public information is managed as a separate verified knowledge base. Connecting the two only when appropriate—without mixing them—is a central principle in GONGJAK's Agent research.
RESEARCH FAQ
Research directions,
expressed as clear answers.
What is the shared foundation of GONGJAK’s research?
Long-term memory that preserves user history and decisions with supporting evidence, AEO that makes company information understandable to both people and AI, and controlled agent architectures that operate only within authorized boundaries.
Why prioritize integration over rebuilding existing systems?
Existing IT tools contain accumulated organizational data and workflows. Adding adapters and an agent layer instead of replacing them entirely can reduce costs, migration time, and operational risk.
BUILD THE NEXT LAYER