GRID v2

General Reconnaissance & Intelligence Dashboard

An all-in-one, local-first agent for OSINT, networks, radio, satellites, microcontrollers, computer vision, and automation — with a persistent, human-like memory layer.

✨ What is GRID?

GRID is a self-aware terminal agent that runs fully on your machine. It combines 16 capability groups — OSINT, network recon, SDR/radio, satellite tracking, microcontroller IoT, computer vision, agent platforms, and automation — under a single conversational interface powered by your local LLM (Ollama), or any OpenAI-compatible backend.

Unlike tool wrappers that live or die on a fixed context window, GRID ships with a layered memory engine inspired by modern agent-memory research (e.g. Tencent's 4-tier memory pyramid). It doesn't forget between sessions: it distills what it learns into reusable knowledge instead of drowning in truncated logs.

🧠 The New: Layered Memory Engine

The latest release adds a real memory system that works across sessions with zero new dependencies (DuckDB keyword recall, fully local).

Layer	What it holds	Where it lives
L3 · Persona	Durable facts & preferences about you	`grid_persona.md`
L2 · Scenarios	Summaries of completed tasks	DuckDB `memory_scenarios`
L1 · Atoms	Standalone factual statements	DuckDB `memory_atoms`
L0 · Transcript	Raw conversation	`memory.md`
Refs	Full verbatim tool outputs (offloaded)	`refs/<id>.md`

How it works:

Every 5 turns, GRID silently runs one LLM distillation pass over recent history via Recaller.distill() — extracting atomic facts, task scenarios, and persona deltas. No feature slice you're still in the middle of gets lost.
On every turn, _build_messages() injects the persona plus only the memories relevant to the current request (keyword recall) alongside your recent turns — so context stays lean and relevant, not bloated.
Tool-output offload: verbose tool results (>800 chars) are written whole to refs/; the model is handed a short preview + a reference id it can pull the full output from on demand — dramatically cutting token usage on long chains.

Memory commands & tools:

Command / Tool	Purpose
`/memory`	Show memory stats (atoms / scenarios / refs / persona)
`/memory <query>`	Recall memories relevant to a query
`/memory clear`	Reset the layered memory store
`/ref <id>`	Read a full offloaded tool output
`memory_recall`	Tool — search layered memory for context
`memory_status`	Tool — memory layer health
`ref_read`	Tool — pull an offloaded ref by id

Inspired by the layered-memory approach popularised by agent frameworks such as Hermes and OpenClaw and TencentDB Agent Memory — humanlike memory, not a bigger scratchpad.

Requirements

Python 3.10+
Ollama (local LLM) — https://ollama.ai
Tesseract OCR (optional) — for image text extraction:
- winget install UB-Mannheim.TesseractOCR
- or https://github.com/UB-Mannheim/tesseract/wiki

Quick Start

# Step 1: Install all dependencies
python install.py

# Step 2: Launch GRID
python grid_agent.py

On first launch, select your LLM backend (Ollama), then pick a model.

What's Included

File	Purpose
`grid_agent.py`	Main app — CLI, tool registry, orchestrator, layered memory
`grid_vision.py`	Computer vision — OCR, face detection, camera validation, video forensics
`grid_osint.py`	OSINT engine — domain/IP/email/phone/username intel, camera search, dorking
`grid_db.py`	DuckDB layer — tool logging, caching, analytics, memory tables
`grid_pb.py`	PocketBase integration — cloud sync, artifact upload
`grid_satellite.py`	Satellite tracking (ISS, passes, TLE, catalog)
`grid_radio.py`	Radio & SDR (Radio-Browser, KiwiSDR, RTL-SDR)
`grid_micro.py`	Microcontroller IoT (ESP32, Arduino, LoRa)
`grid_skills.py`	Self-authored reusable skills
`grid_agent_social.py`	Moltbook social agent platform
`grid_google.py` / `grid_sheets.py`	Google Calendar & Sheets / Excel
`install.py`	One-time setup — deps, models, config
`requirements.txt`	Python package list
`GRID_Tools_Test_Guide.docx`	Full user manual + command reference
`pocketbase.exe`	PocketBase server (optional, for cloud sync)

Screenshots

Capabilities

OSINT — email/phone/username intelligence, IP/domain enrichment, camera search, Google dorking
Network — ping, DNS, netstat, NMAP scanning, netcat suite (listener/client/scan/proxy/transfer/chat)
Computer Vision — OCR, face detection/comparison, camera-stream validation, video forensics
Computer Use — mouse, keyboard, screenshots
Data & Code — CSV/JSON analysis via pandas, arbitrary Python execution
Database — DuckDB SQL queries, PocketBase sync
Memory — persona, facts, scenarios, offloaded refs, relevance ranking, keyword recall
Communication — Telegram, email
System — hardware info, process management, weather
Flight / Satellite / Radio / IoT — live flight tracking, satellite passes, SDR, microcontrollers

Recent Changes — Memorable GRID (v2 Memory)

Added 4-tier layered memory: persona / scenarios / atoms / offloaded refs.
Automatic 5-turn distillation into long-term knowledge via the local LLM.
Relevance-based context injection in place of blind last-5-turn truncation.
Tool-output offload to reduce token usage on long multi-step tasks.
New tools (memory_recall, memory_status, ref_read) and /memory, /ref commands.
Broadened .gitignore to keep local memory/config state out of the repo.

Commands

Command	Action
`/exit`	Save and exit
`/back`	Return to main menu
`/model`	Switch LLM model
`/tools`	View all tools
`/clear`	Clear conversation
`/plan`	Toggle step-by-step planner mode
`/memory [query\|clear]`	View / recall / clear layered memory
`/ref <id>`	Read an offloaded tool output
`/apikey`	Configure API keys (YouTube, Shodan, VirusTotal, etc.)
`/help`	Show command reference

Documentation

Open GRID_Tools_Test_Guide.docx for the complete user manual, tool descriptions, examples, and quick-reference commands.

GridSetup