Software

Run AI agents locally with tool use, memory, and multi-channel messaging. CLI-first, self-hosted.

Multi-agent orchestration framework. Role-based agents with task delegation.

Microsoft's multi-agent conversation framework. Agent-to-agent workflows.

Stateful multi-agent workflows from LangChain. Graph-based agent orchestration.

End-to-end NLP/AI framework by deepset. Pipelines, RAG, agents.

Connect all your devices into an AI cluster. RDMA over Thunderbolt, automatic device discovery, tensor parallelism.

Run 100B+ models at home by joining a distributed swarm. BitTorrent-style collaborative inference.

One-command model runner. Pull and run any GGUF model.

C/C++ inference engine. The foundation most local tools build on.

High-throughput serving with PagedAttention. Production-grade.

Apple Silicon native. Best performance on Mac hardware.

Optimized GPTQ/EXL2 inference for NVIDIA GPUs.

Fast serving framework with RadixAttention. Structured generation.

ChatGPT-style interface for Ollama and OpenAI-compatible APIs.

Desktop app for discovering and running local models. GUI-first.

Offline-first desktop client. Clean UI, local-only by default.

All-in-one RAG application. Documents → local LLM → answers.

Cross-platform desktop client from Nomic. Simple and reliable.

Native Mac/Windows app for running local models. Clean, fast.

2x faster fine-tuning with 80% less VRAM. QLoRA optimized.

Streamlined fine-tuning config. Supports LoRA, QLoRA, full fine-tune.

Hugging Face parameter-efficient fine-tuning. LoRA, prefix tuning, etc.

Transformer Reinforcement Learning. RLHF, DPO, PPO training.

CPU-friendly quantization. Q4_K_M is the sweet spot for most users.

Activation-aware weight quantization. Good GPU inference quality.

GPTQ quantization. Widely supported, GPU-optimized.