Show HN: Core-1 – A distributed C++ AGI framework for trillion-parameter scale

TitanCore released Core-1, a distributed C++ AGI framework supporting up to 1 trillion parameters, featuring a 120-layer Mixture-of-Experts Transformer with full cognitive architecture including persistent memory, reasoning, planning, and online learning across multi-node GPU clusters.

TitanCore Core-1 is a full-stack AGI engine built in C++17 and CUDA . It combines a 120-layer Mixture-of-Experts Transformer with a complete cognitive architecture: persistent memory, structured reasoning, goal-directed planning, meta-learning, world modelling, and continuous online learning — all running across multi-node GPU clusters. | Property | Detail | |---|---| | Version | 1.0.0 | | Release Date | February 2026 | | Status | AGI Framework — Inference-Ready | | Tokenization | Custom BPE — 400,000 token vocabulary | | Weight Format | GGUF titancore.gguf | | Parameters | Up to 1 Trillion | TitanCore Core-1 implements the full Perceive → Remember → Reason → Plan → Act → Learn cognitive loop: ┌──────────────────────────────────────────────────────────────┐ │ TITANCORE AGI COGNITIVE LOOP │ │ │ │ Input │ │ │ │ │ ▼ │ │ ┌──────────────┐ ┌───────────────┐ ┌───────────────┐ │ │ │ Perceive │───▶│ Remember │───▶│ Reason │ │ │ │ Working Mem │ │ Episodic Mem │ │ Chain-of-Thought│ │ │ │ Safety Gate │ │ Semantic Mem │ │ Tree-of-Thought│ │ │ └──────────────┘ └───────────────┘ └───────┬───────┘ │ │ │ │ │ ┌──────────────┐ ┌───────────────┐ ┌───────▼───────┐ │ │ │ Learn │◀───│ Act │◀───│ Plan │ │ │ │ Online GD │ │ Tool Use │ │ MCTS Planner │ │ │ │ EWC + MAML │ │ API Calls │ │ Goal Stack │ │ │ └──────┬───────┘ └───────────────┘ └───────────────┘ │ │ │ │ │ ▼ │ │ ┌──────────────┐ │ │ │ World Model │ Predict future states, detect novelty │ │ │ VAE+Dynamics │ │ │ └──────────────┘ │ └──────────────────────────────────────────────────────────────┘ 120-layer MoE Transformer — 8 experts per layer, top-2 routing FlashAttention v2 — custom CUDA tiled kernel, 128k+ context Paged KV Cache — logical-to-physical block mapping, zero fragmentation RoPE embeddings , SwiGLU MLP, Pre-LayerNorm Parallelism — Tensor ×4, Pipeline ×2, Data ×4, Expert ×8 Learns from every new interaction without forgetting prior knowledge: Online Gradient Descent — real-time weight updates from live data streams Elastic Weight Consolidation EWC — Fisher Information diagonal protects prior knowledge Experience Replay Buffer — 100K capacity, reservoir sampling EMA Weight Snapshots — stable inference weights via exponential moving average Adaptive per-parameter learning rate via AdamW | System | File | Description | |---|---|---| Episodic Memory | core/memory/episodic.cpp | Stores 50K past episodes; cosine-similarity + temporal-decay retrieval | Semantic Memory | core/memory/semantic.cpp | Long-term factual knowledge graph; confidence-scored, conflict-resolved | Working Memory | core/memory/working.cpp | Active context window; attention-weighted importance-based eviction | Four structured reasoning modes: | Mode | Description | |---|---| Standard CoT | Linear step-by-step reasoning with confidence gating | Self-Consistency | Sample N reasoning paths, majority-vote the answer | Tree-of-Thought | BFS branching + value-guided pruning of the reasoning tree | Reflection | Draft → Critique → Revise loop for high-accuracy answers | Monte Carlo Tree Search MCTS with UCB1 selection- Neural-guided rollout policy for state evaluation - Hierarchical goal decomposition into ordered subgoals - Configurable depth, breadth, and exploration constant Learn to learn — adapt to any new task in a few gradient steps: MAML Model-Agnostic Meta-Learning — full second-order FOMAML — first-order approximation faster, production default Reptile — scalable alternative with simple moving-average updates- Fast inference-time adaptation with only a handful of examples Internal predictive model of the environment: VAE Encoder — maps observations to compact latent state z Dynamics Model — predicts next latent z' given z + action Reward Predictor — estimates expected reward from any state Novelty Detection — z-score anomaly flag for unexplored states Imagination — simulate N-step future trajectories for planning Allows the AGI to call external systems: | Built-in Tool | Description | |---|---| calculator | Safe mathematical expression evaluator | web search | Real-time web search via search API | code interpreter | Sandboxed Python execution environment | read file | Secure file system access | db query | Read-only SQL against the knowledge database | Custom tools can be registered at runtime with a schema and handler function. | Component | Minimum | Recommended | |---|---|---| | GPU | NVIDIA A100 80GB ×8 | NVIDIA H100 SXM5 80GB ×8 per node | | Nodes | 1 | 4 32 GPUs total | | System RAM | 512 GB | 1 TB per node | | Interconnect | NVLink | NVLink + InfiniBand 400 Gbps | | Storage | 10 TB NVMe | 100 TB NVMe RAID | | Dependency | Version | |---|---| | OS | Ubuntu 22.04 LTS | | CUDA Toolkit | 12.2+ | | CMake | 3.20+ | | C++ Compiler | GCC 11+ / Clang 14+ | | LibTorch | 2.2+ | | NCCL | 2.18+ | | OpenMPI | 4.1+ | Core-1/ ├── main.cpp AGI master orchestrator ├── CMakeLists.txt │ └── core/ ├── configs/ │ ├── gpt4o.yaml Model & runtime config │ ├── cluster.yaml Cluster topology │ ├── safety.yaml Safety policy │ └── agi.yaml AGI subsystem config │ ├── model/ Transformer backbone ├── distributed/ NCCL, FSDP, MPI ├── optimizer/ ZeRO-3 AdamW ├── dataloader/ Memory-mapped dataset ├── safety/ Moderation, jailbreak, rate limit ├── logging/ Audit trail │ ├── learning/ │ └── online learning.cpp Online GD + EWC + Replay + EMA │ ├── memory/ │ ├── episodic.cpp Past episode store + retrieval │ ├── semantic.cpp Long-term knowledge graph │ └── working.cpp Active context window │ ├── reasoning/ │ ├── chain of thought.cpp CoT / Self-Consistency / ToT / Reflection │ └── planner.cpp MCTS goal-directed planner │ ├── meta/ │ └── maml.cpp MAML / FOMAML / Reptile │ ├── world model/ │ └── world model.cpp VAE encoder + dynamics + reward + novelty │ ├── tools/ │ └── tool executor.cpp Function calling + built-in tools │ └── agi/ └── agi core.cpp Unified AGI cognitive loop controller git clone https://github.com/litonsarkar3988-max/Core-1 cd Core-1 mkdir build && cd build cmake .. \ -DCMAKE BUILD TYPE=Release \ -DTorch DIR=/path/to/libtorch/share/cmake/Torch \ -DCMAKE CUDA ARCHITECTURES="80;86;90" make -j$ nproc ./titancore \ --model core/weights/titancore.gguf \ --config core/configs/gpt4o.yaml \ --agi core/configs/agi.yaml mpirun -np 32 -hostfile hosts.txt \ ./titancore \ --config core/configs/gpt4o.yaml \ --cluster core/configs/cluster.yaml \ --agi core/configs/agi.yaml | File | Purpose | |---|---| core/configs/gpt4o.yaml | Model architecture, quantization, runtime | core/configs/cluster.yaml | Multi-node topology, network, fault tolerance | core/configs/safety.yaml | Content policy, rate limits, PII redaction | core/configs/agi.yaml | All AGI subsystem parameters | All input passes through a mandatory safety pipeline before any model computation: Jailbreak Detection — regex + semantic scan Rate Limiting — sliding-window per user/session Multi-Vector Moderation — embedding-based classifier EWC Knowledge Protection — prevents unsafe fine-tuning from corrupting core knowledge | Phase | Milestone | Status | |---|---|---| | 1 | Core Transformer + CUDA kernels | Complete | | 2 | ZeRO-3 distributed training | Complete | | 3 | Safety & moderation engine | Complete | | 4 | Paged KV cache & inference | Complete | | 5 | Continuous learning Online GD + EWC | Complete | | 6 | Episodic, semantic & working memory | Complete | | 7 | Chain-of-Thought & Tree-of-Thought reasoning | Complete | | 8 | MCTS goal-directed planner | Complete | | 9 | Meta-learning MAML / Reptile | Complete | | 10 | World model VAE + dynamics | Complete | | 11 | Tool use & function calling | Complete | | 12 | Full YAML config parser yaml-cpp | In Progress | | 13 | GGUF weight loader & quantized inference | In Progress | | 14 | 13T token pre-training run | Planned | | 15 | RLHF alignment pipeline | Planned | | 16 | Public API release | Planned | Rahul Sarkar — India GitHub: github.com/Sarkar-AGI https://github.com/Sarkar-AGI Disclaimer:TitanCore Core-1 is an independent research project. NVIDIA GPU hardware is required. CPU execution is not supported.