nouse 0.2.0__tar.gz

nouse-0.2.0/PKG-INFO

@@ -0,0 +1,230 @@
+Metadata-Version: 2.4
+Name: nouse
+Version: 0.2.0
+Summary: The Cognitive Substrate Framework for Model-Agnostic AI — the missing link to AGI
+Author-email: Björn Wikström <bjorn@base76.se>
+License: MIT
+Project-URL: Homepage, https://github.com/base76-research-lab/NoUse
+Project-URL: Repository, https://github.com/base76-research-lab/NoUse
+Keywords: ai,cognition,memory,brain,agi,mcp,plasticity,graph,knowledge-graph
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Developers
+Classifier: Intended Audience :: Science/Research
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.11
+Description-Content-Type: text/markdown
+Requires-Dist: ollama>=0.6.1
+Requires-Dist: rich>=14.3.3
+Requires-Dist: ruamel-yaml>=0.19.1
+Requires-Dist: typer>=0.24.1
+Requires-Dist: kuzu>=0.11.0
+Requires-Dist: watchdog>=4.0.0
+Requires-Dist: numpy>=1.26.0
+Requires-Dist: scipy>=1.12.0
+Requires-Dist: pyvis>=0.3.2
+Requires-Dist: duckduckgo-search>=8.1.1
+Requires-Dist: httpx>=0.28.1
+Requires-Dist: beautifulsoup4>=4.14.3
+Requires-Dist: lxml>=6.0.2
+Requires-Dist: pypdf>=4.2.0
+Requires-Dist: fastapi>=0.135.2
+Requires-Dist: uvicorn>=0.42.0
+Requires-Dist: pydantic>=2.12.5
+Requires-Dist: pandas>=3.0.1
+Requires-Dist: flask>=3.0.0
+Requires-Dist: mcp>=0.1.0
+Provides-Extra: dev
+Requires-Dist: pytest>=9.0.2; extra == "dev"
+Requires-Dist: pytest-asyncio>=1.3.0; extra == "dev"
+Requires-Dist: maturin>=1.12.6; extra == "dev"
+Provides-Extra: brian2
+Requires-Dist: brian2>=2.5.0; extra == "brian2"
+
+<p align="center">
+  <img src="IMG/Nouse.png" alt="Nouse" width="220"/>
+</p>
+
+<h1 align="center">Nouse</h1>
+
+<p align="center">
+  <strong>Persistent domain memory for LLMs. Works with any model.</strong>
+</p>
+
+<p align="center">
+  <a href="https://www.python.org/downloads/"><img src="https://img.shields.io/badge/python-3.11+-blue.svg" alt="Python 3.11+"></a>
+  <a href="https://opensource.org/licenses/MIT"><img src="https://img.shields.io/badge/License-MIT-yellow.svg" alt="MIT"></a>
+  <a href="eval/RESULTS.md"><img src="https://img.shields.io/badge/benchmark-96%25_vs_46%25-brightgreen" alt="Benchmark"></a>
+</p>
+
+---
+
+## The result that motivated this
+
+```
+Model                               Score   Questions
+─────────────────────────────────────────────────────
+llama3.1-8b  (no memory)            46%     60
+llama-3.3-70b  (no memory)          47%     60
+llama3.1-8b  + Nouse memory  →      96%     60
+```
+
+**An 8B model with Nouse outperforms a 70B model without it.**
+
+The effect is not about retrieval. It is about *disambiguation* — a small, precise knowledge signal
+redirects the model's existing priors onto the correct frame. We call this the
+**Intent Disambiguation Effect**.
+
+→ Full benchmark: [eval/RESULTS.md](eval/RESULTS.md)
+
+---
+
+## What Nouse is
+
+Nouse (νοῦς, Gk. *mind*) is a **persistent, self-growing knowledge graph** that attaches to any LLM
+as a memory substrate.
+
+```
+Your documents, conversations, research
+           ↓
+    Nouse knowledge graph
+    (KuzuDB + Hebbian learning)
+           ↓
+    brain.query("your question")
+           ↓
+    Structured context injected into any LLM prompt
+```
+
+It is **not** a RAG system. RAG retrieves chunks. Nouse extracts *relations* — typed, weighted,
+evidence-scored connections between concepts — and injects a compact, structured context block.
+
+It **learns continuously**. Every interaction strengthens or weakens connections (Hebbian plasticity).
+There is no retraining. No gradient descent. The graph grows.
+
+---
+
+## Quick start
+
+```bash
+pip install nouse
+
+# Attach to your knowledge graph
+import nouse
+brain = nouse.attach()
+
+# Query and inject context
+result = brain.query("transformer attention mechanism")
+print(result.context_block())   # inject this into your LLM prompt
+print(result.confidence)        # 0.0 – 1.0
+print(result.strong_axioms())   # verified high-evidence relations
+```
+
+Works with any provider — OpenAI, Anthropic, Groq, Cerebras, Ollama:
+
+```python
+# You handle the LLM call. Nouse handles the memory.
+context = brain.query(user_question).context_block()
+response = openai.chat(messages=[
+    {"role": "system", "content": context},
+    {"role": "user",   "content": user_question},
+])
+```
+
+---
+
+## Run the benchmark yourself
+
+```bash
+git clone https://github.com/base76-research-lab/NoUse
+cd NoUse
+pip install -e .
+
+# Generate questions from your own graph
+python eval/generate_questions.py --n 60
+
+# Run benchmark (requires Cerebras or Groq API key, or use Ollama)
+python eval/run_eval.py \
+  --small cerebras/llama3.1-8b \
+  --large groq/llama-3.3-70b-versatile \
+  --n 60 --no-judge
+```
+
+---
+
+## How the graph grows
+
+```
+Read a document / have a conversation
+           ↓
+    nouse daemon (background)
+           ↓
+    DeepDive: extract concepts + relations
+           ↓
+    Hebbian update: strengthen confirmed paths
+           ↓
+    NightRun: consolidate, prune weak edges
+           ↓
+    Ghost Q (nightly): ask LLM about weak nodes → enrich graph
+```
+
+The daemon runs as a systemd service. It watches your files, chat history,
+browser bookmarks — anything you configure. You never manually curate the graph.
+
+---
+
+## Architecture
+
+```
+nouse/
+├── inject.py          # Public API: attach(), NouseBrain, Axiom, QueryResult
+├── field/
+│   └── surface.py     # KuzuDB graph interface
+├── daemon/
+│   ├── main.py        # Autonomous learning loop
+│   ├── nightrun.py    # Nightly consolidation (9 phases)
+│   ├── node_deepdive.py  # 5-step concept extraction
+│   └── ghost_q.py     # LLM-driven graph enrichment
+└── search/
+    └── escalator.py   # 3-level knowledge escalation
+```
+
+---
+
+## The hypothesis (work in progress)
+
+```
+small model + Nouse[domain]  >  large model without Nouse
+```
+
+We have evidence for this in our benchmark. The next step is to test across
+more domains, more models, and with an LLM judge instead of keyword scoring.
+
+Contributions welcome — especially domain-specific question banks.
+
+---
+
+## Install & run daemon
+
+```bash
+pip install -e ".[dev]"
+
+# Start the learning daemon
+nouse daemon start
+
+# Interactive REPL with memory
+nouse run
+
+# Check graph stats
+nouse status
+```
+
+Requires Python 3.11+. Graph stored in `~/.local/share/nouse/field.kuzu`.
+
+---
+
+## License
+
+MIT — Björn Wikström / Base76 Research Lab

nouse-0.2.0/README.md

@@ -0,0 +1,184 @@
+<p align="center">
+  <img src="IMG/Nouse.png" alt="Nouse" width="220"/>
+</p>
+
+<h1 align="center">Nouse</h1>
+
+<p align="center">
+  <strong>Persistent domain memory for LLMs. Works with any model.</strong>
+</p>
+
+<p align="center">
+  <a href="https://www.python.org/downloads/"><img src="https://img.shields.io/badge/python-3.11+-blue.svg" alt="Python 3.11+"></a>
+  <a href="https://opensource.org/licenses/MIT"><img src="https://img.shields.io/badge/License-MIT-yellow.svg" alt="MIT"></a>
+  <a href="eval/RESULTS.md"><img src="https://img.shields.io/badge/benchmark-96%25_vs_46%25-brightgreen" alt="Benchmark"></a>
+</p>
+
+---
+
+## The result that motivated this
+
+```
+Model                               Score   Questions
+─────────────────────────────────────────────────────
+llama3.1-8b  (no memory)            46%     60
+llama-3.3-70b  (no memory)          47%     60
+llama3.1-8b  + Nouse memory  →      96%     60
+```
+
+**An 8B model with Nouse outperforms a 70B model without it.**
+
+The effect is not about retrieval. It is about *disambiguation* — a small, precise knowledge signal
+redirects the model's existing priors onto the correct frame. We call this the
+**Intent Disambiguation Effect**.
+
+→ Full benchmark: [eval/RESULTS.md](eval/RESULTS.md)
+
+---
+
+## What Nouse is
+
+Nouse (νοῦς, Gk. *mind*) is a **persistent, self-growing knowledge graph** that attaches to any LLM
+as a memory substrate.
+
+```
+Your documents, conversations, research
+           ↓
+    Nouse knowledge graph
+    (KuzuDB + Hebbian learning)
+           ↓
+    brain.query("your question")
+           ↓
+    Structured context injected into any LLM prompt
+```
+
+It is **not** a RAG system. RAG retrieves chunks. Nouse extracts *relations* — typed, weighted,
+evidence-scored connections between concepts — and injects a compact, structured context block.
+
+It **learns continuously**. Every interaction strengthens or weakens connections (Hebbian plasticity).
+There is no retraining. No gradient descent. The graph grows.
+
+---
+
+## Quick start
+
+```bash
+pip install nouse
+
+# Attach to your knowledge graph
+import nouse
+brain = nouse.attach()
+
+# Query and inject context
+result = brain.query("transformer attention mechanism")
+print(result.context_block())   # inject this into your LLM prompt
+print(result.confidence)        # 0.0 – 1.0
+print(result.strong_axioms())   # verified high-evidence relations
+```
+
+Works with any provider — OpenAI, Anthropic, Groq, Cerebras, Ollama:
+
+```python
+# You handle the LLM call. Nouse handles the memory.
+context = brain.query(user_question).context_block()
+response = openai.chat(messages=[
+    {"role": "system", "content": context},
+    {"role": "user",   "content": user_question},
+])
+```
+
+---
+
+## Run the benchmark yourself
+
+```bash
+git clone https://github.com/base76-research-lab/NoUse
+cd NoUse
+pip install -e .
+
+# Generate questions from your own graph
+python eval/generate_questions.py --n 60
+
+# Run benchmark (requires Cerebras or Groq API key, or use Ollama)
+python eval/run_eval.py \
+  --small cerebras/llama3.1-8b \
+  --large groq/llama-3.3-70b-versatile \
+  --n 60 --no-judge
+```
+
+---
+
+## How the graph grows
+
+```
+Read a document / have a conversation
+           ↓
+    nouse daemon (background)
+           ↓
+    DeepDive: extract concepts + relations
+           ↓
+    Hebbian update: strengthen confirmed paths
+           ↓
+    NightRun: consolidate, prune weak edges
+           ↓
+    Ghost Q (nightly): ask LLM about weak nodes → enrich graph
+```
+
+The daemon runs as a systemd service. It watches your files, chat history,
+browser bookmarks — anything you configure. You never manually curate the graph.
+
+---
+
+## Architecture
+
+```
+nouse/
+├── inject.py          # Public API: attach(), NouseBrain, Axiom, QueryResult
+├── field/
+│   └── surface.py     # KuzuDB graph interface
+├── daemon/
+│   ├── main.py        # Autonomous learning loop
+│   ├── nightrun.py    # Nightly consolidation (9 phases)
+│   ├── node_deepdive.py  # 5-step concept extraction
+│   └── ghost_q.py     # LLM-driven graph enrichment
+└── search/
+    └── escalator.py   # 3-level knowledge escalation
+```
+
+---
+
+## The hypothesis (work in progress)
+
+```
+small model + Nouse[domain]  >  large model without Nouse
+```
+
+We have evidence for this in our benchmark. The next step is to test across
+more domains, more models, and with an LLM judge instead of keyword scoring.
+
+Contributions welcome — especially domain-specific question banks.
+
+---
+
+## Install & run daemon
+
+```bash
+pip install -e ".[dev]"
+
+# Start the learning daemon
+nouse daemon start
+
+# Interactive REPL with memory
+nouse run
+
+# Check graph stats
+nouse status
+```
+
+Requires Python 3.11+. Graph stored in `~/.local/share/nouse/field.kuzu`.
+
+---
+
+## License
+
+MIT — Björn Wikström / Base76 Research Lab

nouse-0.2.0/pyproject.toml

@@ -0,0 +1,76 @@
+[project]
+name = "nouse"
+version = "0.2.0"
+description = "The Cognitive Substrate Framework for Model-Agnostic AI — the missing link to AGI"
+readme = "README.md"
+requires-python = ">=3.11"
+license = { text = "MIT" }
+authors = [
+  { name = "Björn Wikström", email = "bjorn@base76.se" }
+]
+keywords = ["ai", "cognition", "memory", "brain", "agi", "mcp", "plasticity", "graph", "knowledge-graph"]
+classifiers = [
+  "Development Status :: 3 - Alpha",
+  "Intended Audience :: Developers",
+  "Intended Audience :: Science/Research",
+  "License :: OSI Approved :: MIT License",
+  "Programming Language :: Python :: 3",
+  "Programming Language :: Python :: 3.11",
+  "Programming Language :: Python :: 3.12",
+  "Topic :: Scientific/Engineering :: Artificial Intelligence",
+]
+dependencies = [
+  "ollama>=0.6.1",
+  "rich>=14.3.3",
+  "ruamel-yaml>=0.19.1",
+  "typer>=0.24.1",
+  "kuzu>=0.11.0",
+  "watchdog>=4.0.0",
+  "numpy>=1.26.0",
+  "scipy>=1.12.0",
+  "pyvis>=0.3.2",
+  "duckduckgo-search>=8.1.1",
+  "httpx>=0.28.1",
+  "beautifulsoup4>=4.14.3",
+  "lxml>=6.0.2",
+  "pypdf>=4.2.0",
+  "fastapi>=0.135.2",
+  "uvicorn>=0.42.0",
+  "pydantic>=2.12.5",
+  "pandas>=3.0.1",
+  "flask>=3.0.0",
+  "mcp>=0.1.0",
+]
+
+[project.optional-dependencies]
+dev = [
+  "pytest>=9.0.2",
+  "pytest-asyncio>=1.3.0",
+  "maturin>=1.12.6",
+]
+brian2 = [
+  "brian2>=2.5.0",
+]
+
+[project.scripts]
+nouse        = "nouse.cli.main:app"
+nouse-brain  = "nouse.daemon.main:main"
+nouse-server = "nouse.web.server:main"
+
+[project.urls]
+Homepage   = "https://github.com/base76-research-lab/NoUse"
+Repository = "https://github.com/base76-research-lab/NoUse"
+
+[build-system]
+requires = ["setuptools>=68", "wheel"]
+build-backend = "setuptools.build_meta"
+
+[tool.setuptools]
+package-dir = {"" = "src"}
+
+[tool.setuptools.packages.find]
+where = ["src"]
+
+[tool.pytest.ini_options]
+asyncio_mode = "auto"
+testpaths    = ["tests"]

nouse-0.2.0/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+

nouse-0.2.0/src/nouse/__init__.py

@@ -0,0 +1,56 @@
+"""
+nouse — The Cognitive Substrate Framework for Model-Agnostic AI.
+
+The missing link to AGI: a persistent, plastic brain layer that gives any LLM
+the cognitive architecture of the human mind.
+
+Memory architecture: working → episodic → semantic → procedural
+Core innovation: KuzuDB knowledge graph + Residual Streams (w, r, u) per edge.
+Plasticity: STDP + Hebbian learning, NightRun consolidation, DeepDive axiom-discovery.
+
+Quick start (high-level kernel API):
+    import nouse
+    k = nouse.Kernel()
+    k.upsert_edge("e1", src="a", rel_type="causes", tgt="b", w=0.3, r=0.0, u=0.6)
+    k.step()
+
+Quick start (knowledge graph API):
+    from nouse.field.surface import FieldSurface
+    field = FieldSurface()
+    field.add_relation("ocean_current", "influences", "climate", why="heat transport")
+"""
+from nouse.kernel import (
+    Brain as Kernel,
+    FieldEvent,
+    NeuromodulatorState,
+    NodeStateSpace,
+    ResidualEdge,
+    MEMORY_TIERS,
+    NEUROMODULATORS,
+    SCHEMA_VERSION,
+)
+
+from nouse.inject import attach, NouseBrain, Axiom, ConceptProfile, QueryResult
+from nouse.search.escalator import EscalationResult
+
+__version__ = "0.2.0"
+
+__all__ = [
+    # Inject API — one-line entry point
+    "attach",
+    "NouseBrain",
+    "Axiom",
+    "ConceptProfile",
+    "QueryResult",
+    # Escalation
+    "EscalationResult",
+    # Residual Stream kernel
+    "Kernel",
+    "FieldEvent",
+    "NeuromodulatorState",
+    "NodeStateSpace",
+    "ResidualEdge",
+    "MEMORY_TIERS",
+    "NEUROMODULATORS",
+    "SCHEMA_VERSION",
+]