Skip to content

MindSpark: ThoughtForge v1.0.0

Choose a tag to compare

View all tags
@hrabanazviking hrabanazviking released this 31 Mar 17:07
· 17 commits to main since this release
v1.0.0
4fd95cf

MindSpark: ThoughtForge v1.0.0

Universal cognitive enhancement layer — offline-first sovereign RAG, TurboQuant inference, and memory-enforced cognition for any model size (1B phone → 70B server).

What Is This?

ThoughtForge is not a model. It is a cognitive layer you wrap around any local GGUF model to give it:

  • Sovereign RAG — offline knowledge retrieval from Wikidata, DBpedia, ConceptNet, GeoNames, and 100+ built-in reference files
  • TurboQuant Inference — multi-draft generation with strict token budget enforcement, auto-detects CUDA/ROCm/Vulkan/Metal/CPU
  • Cognition Scaffolds — intent routing, tone detection, goal-aligned prompt assembly
  • Fragment Salvage — sentence-level scoring + multi-pass refinement to extract the best content from every draft
  • Memory-Enforced Loop — citation integrity, genericness detection, quality gating with repair passes
  • Edge Deployment — Dockerfile, docker-compose, install scripts for Linux/Mac/Windows/Termux/Pi

Hardware Tiers Supported

Profile Target RAM
phone_low Android (Termux) ≤ 2 GB
pi_zero Raspberry Pi Zero ≤ 512 MB
pi_5 Raspberry Pi 5 ≤ 4 GB
desktop_cpu Laptop / desktop CPU ≤ 16 GB
desktop_gpu Gaming GPU (CUDA/Vulkan) ≤ 16 GB VRAM
server_gpu Multi-GPU server 24 GB+ VRAM

Quick Start

git clone https://github.com/hrabanazviking/MindSpark_ThoughtForge
cd MindSpark_ThoughtForge
pip install -e .
python run_thoughtforge.py "What is Yggdrasil?"

Note: query is a positional argument — no --query flag. Omit it entirely for interactive REPL mode.

Test Suite

447 tests passing across 6 phases
python -m pytest tests/ -q

Phases Completed

Phase Name
0 Repo Foundation + Structure
1 Memory Forge + Sovereign RAG
2 TurboQuant Universal Inference Engine
3 Cognition Scaffolds + Orchestration
4 Fragment Salvage + Refinement
5 Edge + Cross-Platform Deployment
6 Testing, Benchmarking, Personality Layer + Release

Benchmarking

from benchmarks.benchmark_profiles import ProfileBenchmark
result = ProfileBenchmark().run("desktop_cpu")
print(result.summary())

Targets: citation accuracy ≥ 85%, enforcement pass rate ≥ 90%.

Persona Consistency Scoring

from benchmarks.persona_consistency import PersonaConsistencyScorer
scorer = PersonaConsistencyScorer()
result = scorer.score(my_response_list)
print(result.summary())   # score ≥ 0.75 = PASS

Built with sovereignty, honor, and a keen edge. ᚠᚢᚦᚨᚱᚲ

Assets 2
Loading