LOADING...
⏸ Development paused · Pivoting to OpenClaw
NodeOne
A locally hosted AI GUI built to explore privacy, agent routing, coding assistance, and memory-aware workflows.
LLaMA-3React / Next.jsPythonSQLiteOllamaLocal Inference
3
Agent Types
2-Stage
Memory Architecture
100%
Local / Private
LLaMA-3
Model Backend
Project Status
Development on NodeOne is temporarily paused. With the creation of OpenClaw, the lessons learned here — particularly around agent routing, persistent memory, local tooling, and privacy-first operation — have been migrated into an in-house OpenClaw infrastructure.
NodeOne remains an important research milestone and a record of the experiments that shaped that system. The project is now being reassessed while its next direction takes shape, rather than actively extended along the original roadmap.
⚡ Why NodeOne?
Cloud AIs are powerful — but they're not private, not always customizable, and not always accessible. NodeOne was built to run entirely on the ForgeCore machine using LLaMA-3, with a flexible GUI that routes between specialized agents, manages persistent tasks, and keeps memory across sessions.
The goal: an AI system that works with you over time — one that remembers context, routes tasks to the right agent, and operates without a subscription or internet dependency.
🧩 Key Features
🖥️
GUI-First Design
Full graphical interface for local AI interaction — no terminal required. Chat-first with agent routing under the hood.
🔀
Agent Routing
Delegate tasks between specialized agents: coding assistant, strategy agent, and utility agent — each with its own context window.
🧠
Two-Stage Memory
Short-term session memory plus long-term persistent memory stored in SQLite. Context survives across restarts.
🔒
Fully Local & Private
All inference runs on-device with LLaMA-3 via Ollama. No data leaves ForgeCore. No API keys, no subscriptions.
⚙️
Modular Architecture
Agent logic in Python, GUI in Next.js. Each component is independently extensible — add new agents or UI panels without touching the core.
🚀
Future-Ready
Designed with multi-modal expansion in mind — vision inputs, tool use, and richer agentic workflows are on the roadmap.
📸 Interface Screenshots
GUI Beta — Main Chat Interface
Two-Stage Memory Architecture
Early NodeOne builds retained as a record of the original research direction
🛠️ Technical Stack
Backend
- ▸ LLaMA-3 via Ollama (local inference)
- ▸ Python agent logic & routing layer
- ▸ SQLite for persistent memory storage
- ▸ REST API between GUI and agents
Frontend
- ▸ React / Next.js GUI
- ▸ TailwindCSS for styling
- ▸ Chat-first interface design
- ▸ Agent selector & task panel
Where the Work Went Next
The project didn't end here.
NodeOne began as an exploration of a local AI-enabled automation node: persistent services, hardware integration, local control, and a physical interface.
As the surrounding AI-agent ecosystem matured, rebuilding every component independently became increasingly difficult to justify. Development shifted from creating another standalone agent runtime toward designing the command and orchestration layer around these systems.
That work evolved into ForgeCommand.
Continue to ForgeCommand →