Neural Lattice · Repos
Code Terminal
Hand-picked open-source AI repos that are actually worth your time. Stars are real (as of Jul 2026). Descriptions cut through the README hype to tell you what each tool actually does and when to use it.
Run LLMs locally with one command
Pull models, serve an OpenAI-compatible API, and manage quantizations — the default local inference layer.
langchain-ai/langchainComposable LLM application framework
Chains, agents, retrievers, and tool integrations — still the widest ecosystem for Python AI apps.
huggingface/transformersPre-trained models for NLP and vision
Load, fine-tune, and deploy thousands of models — the backbone library behind most open-weight workflows.
open-webui/open-webuiSelf-hosted ChatGPT-style UI for Ollama
Multi-user chat, RAG pipelines, and MCP tool routing — the UI layer for basement AI stacks.
ggerganov/llama.cppEfficient LLM inference in C/C++
CPU and GPU inference for GGUF models — what Ollama and LM Studio build on under the hood.
run-llama/llama_indexData framework for LLM applications
Connect private data to LLMs — ingestion, indexing, and query engines with strong RAG primitives.
microsoft/autogenMulti-agent conversation framework
Orchestrate agent teams with human-in-the-loop — Microsoft's take on production agent patterns.
chroma-core/chromaEmbeddings database for AI apps
Lightweight vector store with Python and JS clients — popular choice for local RAG prototypes.
vllm-project/vllmHigh-throughput LLM inference and serving
PagedAttention-based serving engine — the default choice for teams self-hosting models at real production traffic.
BerriAI/litellmUnified API gateway for 100+ LLM providers
One OpenAI-compatible interface for Claude, GPT, Gemini, and local models — swap providers without rewriting app code.
crewAIInc/crewAIRole-based multi-agent orchestration
Lightweight framework for coordinating agent "crews" with defined roles and goals — no LangChain dependency.
qdrant/qdrantVector database written in Rust
Production-grade vector search with filtering — the backbone for larger-scale RAG deployments beyond prototypes.
n8n-io/n8nWorkflow automation with AI superpowers
Fair-code automation tool with native AI agent nodes, LLM integrations, and MCP support. The orchestration backbone behind many self-hosted stacks — including this one.
comfyanonymous/ComfyUINode-based image and video generation UI
Powerful modular interface for Stable Diffusion and FLUX pipelines. Runs on the NAS GPU — drag, connect, generate. The pro's choice for image workflows.
openai/whisperOpen-source speech recognition
Robust transcription in 100+ languages. Runs locally, accurate enough for production, and the foundation for half the voice AI tools built this year.