Awesome MCP › Knowledge & Memory

devflowinc/trieve

⭐ 2717 Rust added to this list on 2025-05-06 repository created 2023-03-26

Trieve is an all-in-one infrastructure platform designed to provide advanced search, recommendations, retrieval-augmented generation (RAG), and analytics through a comprehensive API. It strongly supports integration with the Model Context Protocol (MCP), making it a versatile solution for developers and enterprises looking to enhance their data retrieval and interaction capabilities. The platform offers self-hosting options in virtual private clouds (VPCs) or on-premises environments, with detailed guides for AWS, GCP, Kubernetes, and Docker Compose, ensuring flexibility and control over deployment. Trieve integrates semantic dense vector search by leveraging OpenAI or Jina embedding models combined with Qdrant, enabling high-quality semantic search capabilities. It also supports typo-tolerant full-text and neural search using the SPLADE model, enhancing search accuracy and user experience. The platform features sub-sentence highlighting to emphasize matching words or sentences within search results, improving readability and user engagement. The recommendation system allows finding similar content chunks or grouped files, which is particularly useful for platforms with user interactions like favorites, bookmarks, or upvotes. Trieve's RAG API routes integrate with OpenRouter, providing access to various large language models (LLMs) for retrieval-augmented generation, including fully managed RAG with topic-based memory management and customizable context selection. Users can bring their own models for text embedding, SPLADE, cross-encoder re-ranking, and LLMs, allowing customization and optimization according to specific needs. Hybrid search capabilities combine vector and sparse search with cross-encoder re-ranking for improved result relevance. Additional features include recency biasing to prioritize recent content, tunable merchandising based on user signals, and advanced filtering options such as date-range, substring match, tags, and numeric filters. Grouping functionality enables marking multiple chunks as part of the same file to avoid duplicate top-level search results. The platform is supported by SDKs in TypeScript and Python, comprehensive API documentation, and active community support through Discord and Matrix. Trieve is a robust, flexible, and developer-friendly platform for building advanced search and recommendation systems with strong MCP integration.

https://github.com/devflowinc/trieve

actixactix-webadvanced-searchaianalyticsapiartificial-intelligenceawsbring-your-own-modelscross-encoder-re-rankingdeveloper-friendlydieseldocker-composeembeddingfilteringgcpgroupinghacktoberfesthybrid-searchjinajina-embedding-modelskuberneteslarge-language-modelsllmmcpmodel-context-protocolneural-searchopenaiopenrouterpostgresqlpythonpython-sdkqdrantqdrant-vector-databaseragrecency-biasingrecommendation-apirecommendationsrecommendations-apiretrieval-augmented-generationrustsdksearchsearch-engineself-hostingsemantic-dense-vector-searchsolidjsspladesub-sentence-highlightingtailwindcsstext-embeddingtunable-merchandisingtypescripttypescript-sdktypo-tolerant-searchvector-search