Local AI: 152 Open-Source Tools for 100% Offline LLMs (2025–2026) (github.com)

A community-curated repo published and updated weekly now catalogs 152 open‑source projects to run, fine‑tune, deploy and build with large language models entirely offline for 2025–2026. The collection promises “No cloud · No API keys · No censorship,” and aggregates everything from model runtimes and quantizers to agent frameworks, UIs, vector databases, multimodal stacks and inference servers — a one‑stop toolbox for self‑hosted LLM work aimed at researchers, startups and privacy‑minded teams. Technically, the list spans low‑level runtimes and optimizers (llama.cpp, ExLlama/ExLlamaV2, GGML, vLLM, MLC LLM, ONNX/TensorRT), quantization and fine‑tune tooling (AutoGPTQ, BitsAndBytes, PEFT, TRL, LoRAX), large‑scale training/inference engines (DeepSpeed, Megatron, Colossal‑AI, FSDP), agent/orchestration stacks (LangChain, LlamaIndex, Auto‑GPT, BabyAGI, SuperAGI, Open Interpreter) plus UIs and local servers (text‑generation‑webui, LM Studio, LocalAI, Ollama) and vector stores (Chroma, Qdrant, Milvus, PGVector). It also includes multimodal and audio toolchains (LLaVA, Whisper.cpp, Bark, Coqui) and deployment tooling (Triton, BentoML, Ray Serve). The implications: stronger data privacy, lower recurring costs and vendor independence, but greater ops complexity and hardware needs. The repo is a practical map for anyone building fully offline LLM systems and invites contributions toward 200+ tools.