Repo Radar

Crawl4AI is a web crawler and scraper that converts web content into LLM-ready Markdown, replacing traditional scraping methods.

• drastically more cost-effective than any of the existing solutions

Graphify converts codebases, documentation, and other files into a queryable knowledge graph, replacing traditional code search.

LiteLLM is a Python SDK and proxy server that unifies over 100 LLM APIs into an OpenAI-compatible format.

• 100+ LLMs
• Self-hosted
• Call any LLM in OpenAI format

Headroom is a context compression layer that reduces the token count of inputs before they reach a large language model.

• 60-95% fewer tokens
• 6 algorithms
• reversible

Ponytail is a prompt engineering technique that makes AI agents generate simpler, more concise code by leveraging existing browser capabilities.

• 80-94% less code
• 3-6× faster
• 42-75% cheaper

AirLLM is a tool that optimizes memory for large language model inference, replacing the need for quantization or distillation.

• 70B large language models to run inference on a single 4GB GPU card without quantization, distillation and pruning
• run 405B Llama3.1 on 8GB vram
• Support CPU inference

Context Mode optimizes AI coding agent context windows by replacing raw tool output with sandboxed summaries.

• 98% reduction in context window size from tool output

Supertonic is a text-to-speech system designed for local inference with minimal overhead. It solves the problem of relying on cloud APIs for TTS, enabling on-device processing without privacy concerns

• 31-Language Multilingual
• 99M-Parameter Open-Weight Model
• fast enough to turn an entire webpage into audio in under a second

turbovec is a Rust vector index with Python bindings that replaces FAISS for vector search.

• A 10 million document corpus takes 31 GB of RAM as float32. turbovec fits it in 4 GB
• searches it faster than FAISS
• Hand-written NEON (ARM) and AVX-512BW (x86) kernels beat FAISS IndexPQFastScan by 10–19% on ARM

MiMoCode is a terminal-based AI coding assistant that replaces manual coding and command-line tasks.

• terminal-native AI coding assistant
• persistent memory system to keep a deep understanding of your project across sessions
• continuously improving itself

Semble is a code search library for agents, replacing traditional grep+read methods for code retrieval.

• Uses ~98% fewer tokens than grep+read
• Indexing and searching a full codebase end-to-end takes under a second
• ~200x faster indexing and ~10x faster queries than a code-specialized transformer, at 99% of its retrieval quality

LongCat-Video is a foundational video generation model, replacing traditional methods for creating video content.

Eagle is a vision-language model that serves as a backbone for other models, replacing prior VLM backbones.

• LocateAnything now supports batch inference with a pure FlashAttention runtime
• efficient inference on A100, RTX 4090, and other non-Hopper/Blackwell GPUs

This project offers an AST-based semantic code search engine, replacing traditional keyword or regex-based code search.

• Instant token saving by 70%.
• 1 min setup — install and go, zero config needed!