83,087 visible repositories in the index
49–61 of 61 · Most stars
Run a 105 GB AI model on a Mac that can't hold it. Qwen3.8-Flash-Next 4-bit, streamed from SSD on 16–64 GB Apple Silicon Macs. Fully native Swift, MLX and Metal. Offline, no Python. CLI, Swift library, and Ollama/OpenAI/Anthropic APIs. Works with Claude Code and Codex.
Xolotl: run DeepSeek-V4.1-Flash, all 552B, as a local coding agent on one gaming PC (RTX 4090 + 110 GB RAM): ~33 tok/s on agent tasks, ~41 in fast mode, 512K context. Plus GLM-5.3-Flash, Qwen3.8 and LFM3-Omni. MoE experts tiered across VRAM, pinned RAM and NVMe, with Xolotl Direct NVMe->VRAM. OpenAI + Anthropic API.
DeepSeek-V4.1-Flash-1017 EXL3 on 2x DGX Sparks with TensorFold
TranslatedQ&A on buying, topping up, and top-up-on-behalf for ChatGPT Plus / Pro in China: how much it costs, Alipay, no overseas bank card, new activation or renewal on your own account, upgrading from Plus to Pro100 / 5x before Plus expires, Pro200 / formerly 20x and Pro500 plans, Codex usage, verifying that the payment arrived, and order lookup; includes a runnable offline open-source example.
llama.cpp fork for significantly improved performance on Ampere generation Nvidia cards (especially RTX 3090 / 3090 Ti): example: 110+ tok/s over a 100K-token generation at temperature 1 for qwen3.8, supporting full 262K ctx. Supports low memory (12gb) configs for 200k+ ctx. 120+ TPS, 3300+ prefill for QFN 3.8.
TranslatedGuide to buying and renewing ChatGPT Pro100 (10x, formerly 5x) / Pro200 (20x) in China: Pro100's usage within the Work/Codex plan is doubled with the price unchanged; your own account, Alipay, self-service or manual processing, and a 30-day subscription guarantee. Includes Chinese, English, and Russian.
The repository host gives no description.
PROMETHEVS: run the full DeepSeek-V4.1-Flash-1017 (510 GB) on a gaming PC with 64 GB RAM and a 12 GB GPU. PROMETHEVS engine (compress-then-correct), OpenAI-compatible local server, report and logs.
TranslatedPi.NET is a .NET 10 port of Pi. Pi is a minimal, extensible Agent harness (a TypeScript monorepo of about 171,000 lines). This project ports its core libraries one by one to net10.0, so that the .NET ecosystem can use the same set of capabilities.
TranslatedA mobile web app where the hamster Mugi (むぎ) practices Japanese conversation with you one-on-one. Core idea: the LLM outputs structured replies using a JSON Schema, with the Japanese lines and the Chinese tips in separate fields, and the frontend uses partial JSON parsing to display the reply as it streams in; prompt caching and voice caching are maintained even during long chats. Kanji readings are annotated by the LLM, and any that it misses are filled in by kuromoji on the server side. Corrections are automatically saved into a tips notebook, so you can practice speaking until you have mastered them. For voice, you can choose local VOICEVOX or cloud Google/Azure, and the keys stay only on the local server. Developed in collaboration with Claude Code (Opus 5.5); originally for personal use, it was released as open source with a brand-new git history that contains no personal data or third-party assets. Records of each decision are in docs/decisions/.
Translated大师，我悟了 (Dashi, Wo Wu Le, "Master, I've Got It") | A multi-turn heuristic teaching agent
🧭 Let your coding agent off the leash — not off the rails. Guardrails, a hard budget cap & a self-fixing PR loop for Claude Code / Codex / Gemini. Eval-gated 100/100, you always merge.
Prompt-injection firewall for LLM applications — 33 input detectors, 9 output scanners, federated ed25519-signed threat-intel feed. Apache 2.0, 1040 tests, F1 96.0% with 0% false positives. Docker, GitHub Action, LangChain/LlamaIndex/CrewAI integrations.