Implementation Families+Neural-Symbolic and LLM Integration

Draft — This content has not been approved for publication.

Scope

LLM agent libraries, embedding-based retrieval, and MCP tool integration developed within the Hyperon ecosystem. This family includes both repos with direct MeTTa coupling (metta-motto, petta_lib_localvlm, MeTTa-AI-Assistant) and broader AI application infrastructure built by the same teams (semantic search, MCP chatbots). Pure ML research repos are excluded.

Active Repositories

Core Integration Library

Repo	Language	Upstream	Maturity	Purpose
metta-motto	Python / MeTTa	zarqa-ai	Operational	Primary LLM agent library for MeTTa: OpenAI, Anthropic, OpenRouter integration with stateless wrappers, stateful agents, RAG, SPARQL, functional calling. PyPI: v0.0.12.
petta_lib_localvlm	SWI-Prolog / MeTTa	patham9	Operational	Lightweight Prolog library (~28 lines) for calling local VLM HTTP endpoints (chat completions + embeddings) from PeTTa.

RAG and Semantic Search

Repo	Language	Upstream	Maturity	Purpose
MeTTa-AI-Assistant	Python (FastAPI) + React	iCog-Labs-Dev	Operational	RAG-based MeTTa coding assistant with document/repo ingestion, MeTTa-aware chunking, and chat UI. Docker deployment.
rag-api	Python (Django)	iCog-Labs-Dev	Experimental	Django wrapper around GraphRAG for file upload and global/local retrieval queries. Minimal.
semantic-search-engine	Python (Flask)	iCog-Labs-Dev	Operational	Semantic search API over Slack/Mattermost data using ChromaDB embeddings + Together AI. Has pytest suite.
semantic-search-plugin	React / TypeScript	iCog-Labs-Dev	Operational	Mattermost plugin frontend for the semantic-search-engine backend.
mm-semantic-search	Go + React	iCog-Labs-Dev	Operational	Self-contained Mattermost plugin with Go server + React webapp for semantic search via ChromaDB. GitHub Actions CI.

MCP and Tool Integration

Repo	Language	Upstream	Maturity	Purpose
mcp-xp	Python (FastAPI)	iCog-Labs-Dev	Operational	Chatbot with Galaxy bioinformatics MCP integration. Multi-LLM support (Azure OpenAI, Groq, Gemini). Docker deployment.

How They Fit Together

metta-motto is the central integration layer — the canonical way to call LLMs from MeTTa code. Most other repos in this family either build on top of it or address specific deployment surfaces:

petta_lib_localvlm complements metta-motto for Prolog-side VLM access (local models instead of cloud APIs).
MeTTa-AI-Assistant is a full application using RAG + MeTTa-aware chunking to create a MeTTa coding assistant — the most complete demonstration of neural-symbolic integration in the ecosystem.
The semantic search cluster (semantic-search-engine + plugin + mm-semantic-search) provides embedding-based retrieval for Mattermost — not MeTTa-specific, but developed by the same iCog Labs team as AI integration infrastructure.
mcp-xp demonstrates Model Context Protocol integration with external tool services (Galaxy bioinformatics).

Quick Start

# metta-motto (primary LLM integration)
pip install hyperon-metta-motto
# or from source:
cd metta-motto && pip install -e . && cd tests && pytest

# MeTTa-AI-Assistant (RAG coding assistant, Docker)
cd MeTTa-AI-Assistant/Backend && docker compose build && docker compose up

# semantic-search-engine (embedding search API)
cd semantic-search-engine && pip install -r requirements.txt && python src/server.py

# mcp-xp (Galaxy MCP chatbot)
cd mcp-xp && pip install -r requirements.txt && uvicorn app.main:app --reload --port 8000

Cross-References

metta-nl-corpus (NL-to-MeTTa annotation pipeline) and nl2pln_demo (NL-to-PLN conversion) are covered in Semantic Parsing Full and Reasoning and Search.
metta-motto2 (patham9/metta-motto) is a fork slightly behind the primary zarqa-ai repo. Not listed separately.

Excluded from This Family

ai-dsl: Research archive for AI service composition DSL — legacy, no active development.
FabricPC, PC-Transformers, NGC-PC-Transformers: Pure ML research libraries with no MeTTa/Hyperon coupling.
quantum-neural-architecture: Hybrid quantum-classical ML — no MeTTa integration.

Current State vs. Whitepaper

Outside integration (whitepaper §7.1): metta-motto provides LLM agent wrappers, import hooks, retrieval, and tool calling from MeTTa. The whitepaper's fuller vision of LLMs wrapped as queryable Spaces (matching the Space API) is not yet realized at that abstraction level.
Inside integration / QuantiMORK (whitepaper §7.4): Proposed. No implementation encodes neural structures natively in MORK.
Symbolic Transformer Heads (whitepaper §7.2): Proposed. No implementation augments Transformers with mined AtomSpace templates.

Gaps and Consolidation Opportunities

Three separate semantic search implementations: semantic-search-engine (Python), semantic-search-plugin (React frontend), mm-semantic-search (Go + React, self-contained). The Go version may supersede the Python+React pair.
No MeTTa-native embedding Space: The whitepaper describes Neural Spaces wrapping DNNs as queryable AtomSpaces. No implementation exists.
metta-motto is the single point of LLM access: All MeTTa-LLM integration flows through metta-motto. Its upstream (zarqa-ai) is outside the core Hyperon GitHub organizations.