orbox-brain 1.0.0

ORBOX — Adaptive Unified Runtime Engine for Contextual Intelligence Systems. Memory-native cognitive runtime that sits between you and any LLM.

ORBOX

Adaptive Unified Runtime Engine for Contextual Intelligence Systems

Persistent. Adaptive. Cognitive. Built for frontier intelligence systems.

---

What is ORBOX?

ORBOX is a memory-native cognitive runtime that sits between you and any language model.

It does not generate language — it orchestrates cognition. It remembers, reasons, recalls, and routes — so the model can focus on generation.

   You
    ↓
   ORBOX  →  Memory  →  Intent  →  Routing  →  Validation
    ↓
   Model (Ollama / GPT / Claude / Local)
    ↓
   Validated Output

---

Why ORBOX exists

Frontier language models are powerful but stateless. Every session forgets the last. Every conversation drifts. Every escalation goes unnoticed.

ORBOX is the missing layer:

• Persistent memory across sessions (STM + LTM, JSON-native)
• Emotional and tonal awareness for human-grade interaction
• Adaptive recall — only relevant memory is sent to the model
• Validation layer — blocks drift, hallucination, contradiction, unsafe output
• Model routing — picks the right model for the right intent

---

Built for

• BPO / customer support cognition
• Long-context conversational systems
• Privacy-first AI (runs fully local with Ollama)
• Adaptive runtime research

---

Architecture

orbox/
├── orbox_core.py            Main engine (ORBOXEngine)
├── enhanced_memory.py       Enhanced STM/LTM with embeddings + emotional state
├── embedding_service.py     Embedding generation
├── main.py                  CLI entry
├── core/
│   ├── cognitive_runtime.py     Intent + STM/LTM core
│   ├── model_wrappers.py        Ollama / OpenAI-compatible / Claude wrappers
│   ├── model_router.py          Intent-based routing
│   ├── fallback_manager.py      Failover chain
│   ├── validation_layer.py      Contradiction / hallucination / drift / safety
│   └── ollama_integration.py    Ollama generation
└── run_orbox.sh             Launch script

---

Seven Phases — All Implemented

Phase	Capability	Status
1	Cognitive Runtime Core (STM, intent, weighted recall)	done
2	Semantic Memory Graph (embeddings, relations)	done
3	Frontier Model Wrapper (Ollama / GPT / Claude)	done
4	Tonality + Emotional Runtime (escalation, empathy)	done
5	Adaptive Recall Engine	done
6	Validation Layer (drift, hallucination, safety)	done
7	Long Conversation Stability	done

---

Requirements

• Python 3.10+
• Ollama running locally
• ~2 GB RAM for memory engine
• A local model (e.g. qwen2.5-coder:3b, llama3.2:8b)

---

Quick Start

cd orbox

# Install dependencies
python3 -m venv .venv
.venv/bin/pip install -r requirements.txt

# Make sure Ollama is running
ollama serve

# Pull a model
ollama pull qwen2.5-coder:3b

# Run
./run_orbox.sh

---

Example Interaction

You: my name is Sumit.
ORBOX: Hello Sumit. I will remember you across sessions.

[session ends — start a new one]

You: do you remember me?
ORBOX: Yes — your name is Sumit. We discussed cognitive runtime
       architecture in our last session.

---

What ORBOX is NOT

• Not a chatbot wrapper
• Not a fine-tuned model
• Not open-source — see LICENSE
• Not free to copy, fork, or commercialize

---

License

ORBOX is proprietary software.

See LICENSE for the full terms.

In short:

• You may view the source and run it locally for personal evaluation.
• You may not copy, modify, distribute, reimplement, benchmark, or build competing products.
• All techniques, formulas, and architectural patterns are reserved.

For commercial licensing or collaboration: [e-mail]

---

Author

Sumit Verma Independent researcher — cognitive runtime systems BPO domain expertise meets AI architecture.

---

ORBOX — Where context evolves into adaptive intelligence.