Flatseek # Flatseek ### Open Knowledge Infrastructure for AI **Search. Retrieve. Run. Train.** Open-source infrastructure for building search systems, AI applications, and language models from your own knowledge.
--- ## About Flatseek is an open-source ecosystem for modern AI infrastructure. Instead of treating search, retrieval, model training, inference, and model storage as isolated problems, the ecosystem provides modular projects that work independently or together as a complete AI pipeline. Index your knowledge once, then reuse it across search, semantic retrieval, Retrieval-Augmented Generation (RAG), model training, inference, analytics, and deployment. ```text Knowledge │ ┌───────────────────┴───────────────────┐ │ │ ▼ ▼ Flatseek Flatvec Keyword Search Semantic Search │ │ └───────────────────┬───────────────────┘ │ ▼ Flatask RAG Runtime │ ┌───────────────────┴───────────────────┐ │ │ ▼ ▼ Flattune Flatbuild Dataset Generation Train Models & Fine-tuning From Scratch │ │ └───────────────────┬───────────────────┘ │ ▼ Flatweight Model Storage Layer │ ▼ Flatrun Streaming Inference │ ▼ Flatlens Analytics & Exploration ``` --- ## Projects | Project | Description | |---------|-------------| | **[Flatseek](https://github.com/flatseek/flatseek)** | Disk-first full-text search engine for portable datasets and production-scale search. An alternative to Elasticsearch, OpenSearch, and Meilisearch. | | **[Flatvec](https://github.com/flatseek/flatvec)** | Disk-first vector search engine for semantic retrieval and RAG. An alternative to FAISS, Milvus, Pinecone, Qdrant, and Chroma. | | **[Flatask](https://github.com/flatseek/flatask)** | Retrieval-Augmented Generation (RAG) runtime that connects keyword search, semantic search, and language models into a unified retrieval pipeline. | | **[Flattune](https://github.com/flatseek/flattune)** | Dataset generation and fine-tuning pipeline for creating high-quality training corpora from existing knowledge. | | **[Flatbuild](https://github.com/flatseek/flatbuild)** | Framework for training transformer language models from scratch, including tokenizer training, chat templates, checkpoint export, and GGUF conversion. | | **[Flatweight](https://github.com/flatseek/flatweight)** | Storage-native model format and tooling for organizing, converting, and managing model weights for efficient deployment and streaming inference. | | **[Flatrun](https://github.com/flatseek/flatrun)** | Streaming inference runtime that executes GGUF, SafeTensors, and MLX models directly from storage with minimal memory usage. | | **[Flatlens](https://github.com/flatseek/flatlens)** | Analytics and visualization interface for exploring search indexes, datasets, retrieval pipelines, and AI workflows. | --- # Design Principles The Flat ecosystem is built around a few simple ideas: - Build knowledge once and reuse it everywhere. - Keep data portable and storage-native. - Prefer open formats over proprietary services. - Make every project useful on its own. - Compose projects into larger AI workflows when needed. - Train, deploy, and share models with reproducible pipelines. - Run efficiently on consumer hardware whenever possible. --- # License Unless otherwise noted, projects and published artifacts are released under the Apache License 2.0. ---
### Build once. Search anywhere. Train from scratch. Run everywhere.