AI Coding2026
Maps a codebase plus docs, PDFs, media and configs into a local, queryable knowledge graph; parses code with a local tree-sitter AST (no LLM), uses configurable backends for semantic extraction of non-code, and outputs graph.json, graph.html and a brief report.
AI Client2026
Compresses LLM/agent replies into a terse “caveman” style to cut output tokens (~65–75%) while preserving technical accuracy. Offers per-agent skills, intensity modes, memory-compression and middleware to lower token cost and extend usable context.
AI Model2026
Provides a compact GGUF export of a tuned Gemma‑4 26B variant for local inference, optimized for llama.cpp and Apple Silicon to deliver faster, less‑censored chat and coding outputs. Includes Q4_K_M quantization and a neutral embedded template for more reliable local deployments.
AI Coding2026
Turns a codebase into a live structural knowledge graph that coding agents can query in milliseconds. Bi-temporal, replay-aware indexing of symbols and relationships performed locally with zero LLM API calls; Rust-native, MCP-native integrations and fast incremental updates.
AI Model2026
A Mixture-of-Experts instruct-capable LLM (295B total, 21B active) designed for long-context reasoning, code/agent workflows and instruction-following; released by Tencent Hy Team with safetensors weights on Hugging Face.
AI Model2026
Multimodal agent model for long-horizon coding, image-text understanding, and autonomous task orchestration. Built as a 1T-parameter Mixture-of-Experts with 256K context and native int4 quantization — intended for coding-driven design, persistent background agents, and swarm-style sub-agent workflows.
AI Model2026
A healed 64-layer 'frankenmerge' that stacks two Qwen3.5-derived finetunes into an ~18B GGUF model for multilingual text generation, reasoning, and reliable code/frontend output. Healed with a 1000-step QLoRA to reduce layer-boundary artifacts and targeted to run on 12–16 GB GPUs.
AI Dataset2026
Provides 34k execution-style agent trajectories (11,766 issues) for supervised fine-tuning of code-focused LLMs. Each instance includes multi-step interactions, tool-call records, and final unified diffs; generated with Qwen3-Coder and released under permissive licenses for commercial use.
AI Dataset2026
Provides 316,427 Go source-code samples in JSONL focused on concurrency and backend idioms, enabling fine-tuning and evaluation of code models for completion, summarization, and static-analysis tasks.
AI Model2026
An ~18B frankenmerge text-generation model that stacks two 32-layer Qwen3.5-based finetunes and ships as a 9.2GB Q4_K_M GGUF for efficient local inference. A 1000-step QLoRA heal reduces layer-boundary code corruption and targets coding, reasoning, multilingual chat, and 12–16GB GPU compatibility.
AI Dataset2026
Synthetic JSON dataset of model-generated prompts and step-by-step reasoning traces (≈90k rows, ~75M tokens) created with Claude Sonnet 4.6 and cross-checked by Gemini 3.1 Pro — intended for training or fine-tuning LLMs on natural reasoning, multi-domain code/math, and instruction following. Hosted on Hugging Face, MIT license.
AI Model2026
A 27B multimodal causal language model with a vision encoder and native long-context support (262,144 tokens). Optimized for repository-level coding agents and multimodal understanding; includes preserved "thinking" traces, multi-token prediction (MTP), and deployment recipes for vLLM / SGLang / Transformers.