GitHub Repositories

Draft — This content has not been approved for publication.

← Back to Reference

A comprehensive index of GitHub repositories in the Hyperon ecosystem, organized by role and maturity. Covers the 5 primary GitHub organizations and 200+ local checkouts in the reference collection.

This page is the GitHub-facing repository guide for readers who need a fast orientation to the code surface. For the more curated architectural interpretation layer, see Implementation Families. Mirrors, snapshots, and non-GitHub surfaces are intentionally not all listed individually here.

Investor / First-Read Shortlist

RepositoryWhy it mattersRole
hyperon-experimentalMain reference implementation of the MeTTa language and runtime surface.Core language runtime
MORKMost distinctive new substrate work: triemap storage, zipper VM, and reduction kernel experiments.Knowledge substrate
dasCloud-native distributed memory layer and query surface for Hyperon-era systems.Distributed storage
PLNCurrent probabilistic reasoning implementation and rule surface.Reasoning engine
asi-chainOn-chain execution and alliance-facing runtime layer.Blockchain runtime
MeTTa-CompilerPure Rust MeTTaTron evaluator and the strongest current compilation path into the F1R3FLY stack.Compiler / runtime
metta-mosesMain modern MOSES-in-MeTTa line rather than the older C++ codebase.Evolutionary learning
custom-atomspace-builderConcrete domain-facing pipeline showing how Hyperon-style knowledge substrates get used in practice.Application pipeline

At a Glance

SystemPrimary RepositoryRoleStatus
MeTTa Languagehyperon-experimentalCore logic engine; Python, C, and Rust APIsStable
Hypergraph KernelMORKHigh-performance triemap substrate for MeTTaActive Research
Distributed MemorydasCloud-native distributed knowledge graphActive Research
Inference EnginePLNProbabilistic Logic Networks for uncertain reasoningActive Research
Blockchain Runtimeasi-chainASI Alliance on-chain AI execution layerActive Research
Legacy AtomSpaceatomspacePredecessor hypergraph DB; widely deployedMaintained

Organizations

OrganizationFocusRepos
trueagi-ioHyperon core: MeTTa, MORK, PLN, Pattern Mining18
opencogAtomSpace, cogutil, NLP, legacy infrastructure40+
singnetDAS, AIRIS, AI-DSL, SingularityNET services30+
iCog-Labs-DevMOSES, ECAN, bioinformatics, search, MORK SDKs49+
F1R3FLY-ioMeTTaTron, MeTTa-IL, Rholang runtime10+

Core Runtimes

MeTTa language implementations and evaluation engines.

  • trueagi-io/hyperon-experimental β€” Official Rust reference implementation (v0.2.10). Multi-crate workspace with Python/C bindings. Stable.
  • patham9/PeTTa β€” High-performance Prolog-based compiler with Smart Dispatch. Active.
  • F1R3FLY-io/MeTTa-Compiler (MeTTaTron) β€” Rust runtime targeting MeTTa-on-Rholang. Tree-sitter parser; iterative trampoline tree-walk interpreter (not a code generator at HEAD); MORK/PathMap retained for rule storage + pattern-match, HashMap for multiset set-ops (3.5–5.3Γ— speedup over PathMap per the 2025-12-29 benchmark). Rule matching filters to most-specific rules β€” divergent from hyperon-experimental v0.2.10. Direct-Rust Rholang integration via shared f1r3node/models protobuf types (no FFI). At HEAD efee4be (2026-01-19, Cargo v0.2.0, 190 commits ahead of the v0.2.0 Git tag), main is 4 months stagnant but feature branches (feature/wam-backend, feature/pln-support, feat/bind-tokenizer) remained active through 2026-03. Per MeTTa runtime cluster pilot Source 3, closed 2026-05-13: [METTATRON-LAZY-NONDETERMINISTIC-CARTESIAN] / [TREE-WALK-INTERPRETER] / [PARSER-TREE-SITTER] / [STDLIB-EXTENDED-WITH-SET-OPS] / [RHOLANG-DIRECT-RUST-INTEGRATION] / [MORK-PATHMAP-HYBRID-PER-OPERATION] / [ASYNC-PER-EXPRESSION-PARALLEL] / [Q-3-DIVERGENT-SEMANTICS]. Active on branches; main stagnant.
  • trueagi-io/metta-wam (MeTTaLog) β€” Warren Abstract Machine interpreter/transpiler with REPL and LSP. Active.
  • trueagi-io/jetta (JeTTa) β€” JVM/Kotlin compiler with ANTLR parsing and ASM bytecode generation. Active.
  • trueagi-io/metta-morph β€” Macro-based MeTTa-to-Chicken-Scheme translator (10–200x speedup). Active.
  • F1R3FLY-io/MeTTaIL β€” Scala 3 + BNFC + Haskell intermediate language interpreter (GSLTs). Active.
  • trueagi-io/FormalMeTTa β€” Literal Scala implementation of MeTTa specification. Reference.
  • opencog/atomspace-metta β€” C++ ForeignAST bridge from MeTTa to AtomSpace. Active.

Cognitive Logic & Reasoning

Inference engines, learning algorithms, and autonomous agent reasoning systems.

  • trueagi-io/PLN β€” Probabilistic Logic Networks (consolidated lib_pln.metta). Active.
  • trueagi-io/pln-experimental β€” Experimental PLN extensions.
  • trueagi-io/chaining β€” Forward/backward chaining, program synthesis, metamath proofs. Active.
  • opencog/miner β€” Legacy symbolic pattern miner (C++/Guile Scheme); URE-based; "Frequent and surprising subhypergraph pattern miner for the AtomSpace". The canonical baseline that the hyperon-miner trio MeTTa-ports. Per Perception cluster pilot Source 2 (2026-05-01): paper-faithful to the Chi/Xia/Yang/Muntz 2005 methodology family. ngeiswei is the upstream maintainer; per Non-clustered HAA cluster pilot Source 3 (2026-05-06), the leungmanhin/singnet/ngeiswei forks are all stale or merged-behind. Maintained.
  • trueagi-io/hyperon-miner β€” MeTTa port of opencog/miner. Pattern mining for frequent/surprising hypergraph patterns. iCog-Labs-Dev/hyperon-miner is the active fork at HEAD; zariuq/hyperon-miner is Zar Goertzel's own fork. Active.
  • rejuve-bio/neural-subgraph-matcher-miner β€” GNN-based neural subgraph pattern miner; SPMiner-lineage; paradigm-distinct from symbolic pattern mining. Per Perception cluster pilot 2026-05-01: [GNN-NEURAL-MINING] / [PARALLEL-NON-INTEGRATED] with hyperon-miner-family symbolic miners. See Neural Pattern Mining. Active.
  • iCog-Labs-Dev/conceptBlending β€” Concept blending implementation repo with four parallel sub-implementations: an_infotheoretic_approach/ (info-theoretic with GA + CMA-ES), peTTa_version/ (PeTTa port with multi-objective CMA-ES), a_categorytheoretic_approach/ (category-theoretic colimit/pushout), a_naive_approach/ (graph matching + GPT prototype). Per Non-clustered HAA cluster pilot Source 4 close 2026-05-07: [CONCEPT-BLENDING-PARTIAL-IMPLEMENTATION] β€” info-theoretic + GA/CMA + category-theoretic colimit branches are [IMPLEMENTATION-BACKED-CORE] at HEAD 9675e5e; quantale + paraconsistent + HoTT-adjacent material is [PAPER-ONLY-FORMAL-EXTENSION]. See Concept Blending. Active.
  • iCog-Labs-Dev/metta-moses β€” Full MeTTa reimplementation of MOSES evolutionary learning. Per Non-clustered HAA cluster pilot Source 5 close 2026-05-07: [ACTIVE-METTA-PRIMARY] at HEAD 9f69e17 (2026-03-26); the canonical Hyperon-era MOSES line. See MOSES. Active.
  • opencog/asmoses β€” AtomSpace-MOSES; the active C++ AtomSpace MOSES baseline. Per HAA Source 5: [ACTIVE-CPP-ATOMSPACE-BASELINE] at HEAD dde4849c4 (2025-12-10). linas/as-moses is byte-identical at the same SHA [VEPSTAS-MIRROR-SAME-SHA] β€” not a separate implementation tradition. Maintained.
  • trueagi-io/hyperon-moses β€” Hyperon-era MOSES boolean-reduction prototype. Per HAA Source 5: [SUPERSEDED-BY-METTA-MOSES] for general MOSES; retained as boolean-reduction / ENF-pipeline prototype. P12-verified 27 commits ahead of ngeiswei/hyperon-moses older sibling. Active.
  • opencog/moses β€” Legacy C++ MOSES implementation (the 2008-era pre-AtomSpace original; README explicitly notes "kept to guaranty backward compatibility" and points to AS-MOSES). Per HAA Source 5: [ARCHIVAL-PARADIGM-PREDECESSOR] β€” last commit 2023-09-19. Archival.
  • MOSES stale lineage (HAA Source 5 V5-9 lump; orchestrator-verified P12 verbatim counts): leungmanhin/asmoses [STRICT-FORK-STALE] (0 ahead / 107 behind); singnet/asmoses [STRICT-FORK-STALE-MERGED-UPSTREAM] (7 ahead = merge-sync metadata / 165 behind); singnet/moses [STRICT-FORK-DIVERGED-LEGACY-FIXES] (19 ahead / 23 behind). None contribute new traditions. iCog-Labs-Dev/moses-optimization-algorithms is [STANDALONE-EXPERIMENT-NOT-FORK] β€” independent commit history (Initial commit / hill climbing / simulated annealing / Nahom32 PR), not derived from asmoses or moses.
  • iCog-Labs-Dev/metta-attention β€” MeTTa-based ECAN with mind agents. Active.
  • opencog/attention β€” Legacy OpenCog ECAN (C++). Archival.
  • iCog-Labs-Dev/hyperon-openpsi β€” OpenPsi motivational framework substrate (MeTTa); the operational predecessor of MetaMo. Per Non-clustered HAA cluster pilot Source 2 close 2026-05-06: [OPENPSI-PREDECESSOR-SUBSTRATE] β€” at HEAD 3b356c5 contains zero MetaMo/MAGUS naming, so this is NOT a MetaMo strict-literal implementation; the appraisal comonad in MetaMo abstracts this substrate at one level higher. See MetaMo. Active.
  • glicerico/MetaMo-Prototype β€” Heuristic LangGraph/LLM rewrite of motivational dynamics; Python; forked from dagemawinegash/MetaMo-Prototype. Per Non-clustered HAA cluster pilot Source 2 close 2026-05-06: [HEURISTIC-PROTOTYPE] β€” no formal categorical structure; targeted searches at HEAD for "bimonad" / "comonad" / "MAGUS" / "OpenPsi" return zero matches. Parallel-paradigm to MetaMo-Python. Active.
  • patham9/metta-nars β€” Non-Axiomatic Reasoning System in MeTTa. Per Non-clustered HAA cluster pilot Source 5 close 2026-05-07: [IMPLEMENTATION-BACKED-NAL1-5] / [METTA-NARS-NOT-PLN] at HEAD 945e2f8 (2024-10-14); ONA-style NARS port with f/c truth values. Paradigm-distinct from PLN β€” cross-link as boundary, not as alternative implementation. See MeTTa-NARS. Active.
  • patham9/NACE β€” Non-Axiomatic Causal Explorer. Per Non-clustered HAA cluster pilot Source 5 close 2026-05-07: [IMPLEMENTATION-BACKED-CORE] / [AIRIS-DERIVED-NAL-ADJACENT] / [NACE-NOT-AIRIS] at HEAD 361fddf (2025-05-06); pure-Python (no src/ or C++; PyQt GUI in gui.py); extends AIRIS apparatus with partial observability + NAL evidence; bridges to MeTTa-NARS via bridge.py. The "1000x data efficiency" framing should be paper-sourced if retained. See NACE. Active.
  • berickcook/AIRIS_Public β€” Autonomous Intelligent Reinforcement Inferred Symbolism (Cook & Hammer 2024). Cook's reference / stable builds; the paper-cited canonical implementation (Cook & Hammer 2024 footnote 3). Per Non-clustered HAA cluster pilot 2026-05-06: previously misattributed here to singnet/AIRIS_Public (HTTP 404); URL corrected at cluster-close. Active.
  • singnet/AIRIS-general β€” Cook's SingularityNET-side generalized rewrite of AIRIS; pure-Python single-file library accepting structured/unstructured JSON inputs. MIT. Active.
  • singnet/ai-dsl β€” AI Domain Specific Language for composing AI services. Per Non-clustered HAA cluster pilot Source 5 close 2026-05-07: [ACTIVE-IDRIS-DSL] / [AI-DSL-DUAL-CITIZEN] at HEAD aaa702e (2025-06-25); two implementation tracks β€” Idris dependent-types layer (77 .idr files in experimental/ai-algorithms/) and MeTTa combinator/ontology layer (30+ .metta files in experimental/ai-service-composition/). Center of gravity is service composition for the SingularityNET marketplace; the MeTTa combinator track is the HAA-relevant slice. NOT a core cognitive algorithm comparable to PLN/ECAN/MOSES/NARS. See AI-DSL. Active.
  • Semantic parsing repo trilateral (per Non-clustered HAA cluster pilot Source 4 close 2026-05-07; previously cited singnet/semantic-parsing here returned HTTP 404 and was corrected to this trilateral set):
    • rTreutlein/semantic_parsing β€” Canonical upstream; Python LLM-driven NL β†’ predicate logic / MeTTa pipeline. Active.
    • leungmanhin/semantic_parsing β€” Active fork-derived primary; primary committers Leung Man Hin and Roman Treutlein. Active.
    • leungmanhin/semantic-parser β€” Separate dashed-name Hyperon/PeTTaChainer line (no upstream remote configured); PLN expression guidance + Premises/Conclusions syntax. Active.
    Per HAA Source 4: [SEMANTIC-PARSING-SPLIT-STACK] β€” these are LLM-to-PLN/MeTTa, NOT wrappers around classical Link Grammar / RelEx (which see under Natural Language Processing). See Semantic Parsing.
  • zarqa-ai/metta-motto β€” LLM agent integration library (OpenAI, Anthropic, OpenRouter). Active.

Infrastructure & Storage

Hypergraph databases, storage backends, and distributed data infrastructure.

  • trueagi-io/MORK β€” MeTTa Optimal Reduction Kernel. Triemap DB + zipper VM. Nightly Rust. Active research.
  • opencog/atomspace β€” OpenCog AtomSpace hypergraph database (C++/Scheme/Python). Maintained.
  • singnet/das β€” Distributed AtomSpace with Bazel build and multiple backends. Active.
  • singnet/das-metta-parser β€” C Flex/Bison MeTTa parser for DAS (MongoDB/Redis).
  • singnet/das-toolbox β€” CLI tools for DAS infrastructure.
  • opencog/atomspace-rocks β€” RocksDB persistent backend for AtomSpace; per AtomSpace cluster pilot 2026-04-29, this is the Phase 5+ primary durability candidate for Hyperon, with MORK-PATH as secondary (delete-blocker) and DAS-PATH for Phase 6+ distribution. Maintained.
  • opencog/atomspace-cog β€” Distributed AtomSpace network client.
  • opencog/atomspace-storage β€” Base StorageNode API and I/O subsystems.
  • opencog/atomspace-bridge β€” SQL-to-AtomSpace read/write bridge.
  • singnet/pattern-index (forked from opencog/pattern-index) β€” Disk-based AtomSpace pattern indexing with C++/Scheme APIs; "an index for fast lookup of patterns in the AtomSpace". Has its own pattern-mining algorithms but is NOT a wrapper around the PatternMatcher or PatternMiner. Per Non-clustered HAA cluster pilot Source 3 close 2026-05-06: [PATTERN-INDEX-SEPARATE-PARADIGM-ADJACENT] β€” adjacent legacy infrastructure to Pattern Mining, distinguished by RAM-only AtomSpace mining (the symbolic paradigm) vs disk-based subgraph indexing (this repo). Maintained / Archival.
  • Adam-Vandervorst/CZ2 β€” Scala 3 triemap toolkit and MORK scaling experiments.
  • iCog-Labs-Dev/mork-rust-sdk β€” Rust SDK for MORK API.
  • iCog-Labs-Dev/mork-ts-sdk β€” TypeScript SDK for MORK HTTP API.

Distributed AI & Blockchain

On-chain execution environments and decentralized AI infrastructure.

Applied AI & Vertical Applications

Domain-specific applications of Hyperon capabilities in longevity research, bioinformatics, and enterprise AI.

Simulated Environments & Agents

Agent frameworks, virtual environments, and embodied AI testbeds.

  • trueagi-io/Vereya β€” Minecraft 1.21 Fabric mod for AI agent control (Java 21). Active.
  • trueagi-io/minecraft-demo β€” Python API for Minecraft AI agents via Vereya. Active.
  • opencog/rocca β€” Rational OpenCog Controlled Agent (OpenAI Gym). Research.
  • opencog/sensory β€” Low-level sensory I/O Atoms (C++). Part of Linas Vepstas's Hyperon-era AtomSpace-Scheme research portfolio (per Perception cluster pilot 2026-05-01); 7 wired sensory atom types incl. OllamaNode dual-characterization. See Sensory. Active research (status updated from "Archival" per Perception pilot).
  • opencog/agents β€” Agent atoms; Vepstas Hyperon-era research portfolio sibling to opencog/sensory. Active research.
  • opencog/motor β€” Motor-output atoms; Vepstas Hyperon-era research portfolio sibling. Active research.
  • opencog/evidence β€” Evidence-tracking atoms; Vepstas Hyperon-era research portfolio sibling. Active research.
  • opencog/atomese-simd β€” SIMD-vector operations on AtomSpace; Vepstas Hyperon-era research portfolio sibling. Active research.
  • opencog/vision β€” Vision atoms scaffolding (6 wired types: ImageNode/ImageValue + 4 ImageLink ops). Per Perception cluster pilot 2026-05-01: [SCAFFOLDING-NOT-PIPELINE]. See Vision. Scaffolding.
  • opencog/spacetime β€” 4D spacetime tracking in AtomSpace (C++). Archival.
  • opencog/ros-behavior-scripting β€” Hanson Robotics Eva API (Python/Scheme). Archival.
  • opencog/ghost_bridge β€” ROS bridge for GHOST chatbot system. Archival.
  • patham9/nartech_ros β€” NARS + ROS robot integration. Active.

Developer Ecosystem

Tools, training materials, IDE support, and integration utilities for MeTTa developers.

Natural Language Processing

  • opencog/link-grammar β€” CMU Link Grammar parser (C). v5.13.1. Active maintenance per Non-clustered HAA cluster pilot Source 4 close 2026-05-07. Maintained.
  • opencog/relex β€” English dependency relationship extractor (Java). v1.6.3 circa 2016. Per HAA Source 4: maintained-legacy-dependency-parser; recent commits are dependabot maven dependency bumps, NOT algorithmic modernization. Archival.
  • opencog/learn β€” Neuro-symbolic language learning (Scheme); Atomese LG-based parsing framework with MST/MPG parses; recent neuro-LG.lyx diary. Active research.
  • opencog/generate β€” Network generation from syntax (C++); AtomSpace-based lexical entries. Active.
  • opencog/lg-atomese β€” Atomese API for Link Grammar.
  • opencog/matrix β€” AtomSpace sparse vector/matrix library (Scheme).
  • rTreutlein/nl2pln_demo β€” Natural language to PLN conversion demo (2 commits, demo-only). Per HAA Source 4: [DEMO-NOT-PIPELINE].

Knowledge Converters & Pipelines

Visualization

Research & Experimental

Legacy OpenCog Infrastructure

Core C++ framework components. Maintained for AtomSpace compatibility; not the Hyperon-forward path.

Community Resources


Related cards: Implementation Families Β· Technical Companions Β· TrueAGI Β· iCog Labs



Discussion