[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$f1lv25gug2se42":3},{"code":4,"message":5,"data":6,"timestamp":58},10000,"success",{"full_name":7,"position":8,"description":9,"language":10,"stars":11,"forks":12,"stars_gained":13,"age_days":14,"hn_threads":15,"one_liner":22,"summary_md":23,"topic":24,"github_topics":27,"license":48,"homepage":49,"updated_at":50,"history":51},"Tencent\u002FWeKnora",16,"Open-source LLM knowledge platform: turn raw documents into a queryable RAG, an autonomous reasoning agent, and a self-maintaining Wiki.","Go",35330,4755,94,459,[16],{"id":17,"title":18,"points":19,"comments":20,"posted_at":21},46266357,"WeKnora – LLM-Powered Document Understanding and Retrieval Framework",2,0,"2025-12-14","Turns a team's scattered documents into a searchable RAG, a tool-using agent and a self-maintained wiki on one platform.","WeKnora is worth a look if you want one self-hosted place where a document collection becomes Q&A with citations, a multi-step agent, and a browsable wiki. Treat it as a platform to operate, not a library to drop in.\n\n## The problem\nA team already has PDFs, Word files, meeting notes and pages spread across Feishu, Confluence or GitLab. Answering a question today means searching several systems, opening documents and re-reading them, and knowledge that gets recombined into new notes usually lives in someone's chat history. WeKnora ingests those sources into knowledge bases, so a question is answered from retrieved chunks with citations, and the agent can chain retrieval with web search, MCP tools and a sandbox to finish a task. The new part is that the same bases also feed an agent-generated wiki, so the output of one workflow becomes material for the next.\n\n## What's different\nThe design puts three products on one set of knowledge bases: quick RAG-style Q&A, a ReAct agent with tools, and an interlinked wiki with a knowledge graph. That matters because a single ingestion and chunking pipeline serves all three, instead of stitching together a RAG stack and a separate agent framework. The README also documents chunk-level editing with revisions, per-upload parsing and chunking settings, and session-persistent Docker\u002FE2B\u002FCube sandboxes with network policy.\n\nThe trade-off is breadth. Supporting many vector databases, object stores, data sources, IM channels and 29 model vendors means plenty of configuration combinations, and the operational surface grows with each optional profile. Running with `docker compose` starts core services, but Neo4j for the graph, MinIO for object storage and Langfuse for tracing are separate profiles you enable deliberately. Breadth here is a menu, not a promise that every combination is exercised; check the [feature docs](https:\u002F\u002Fweknora.weixin.qq.com\u002Fdocs\u002F03-features\u002F02-knowledge-base) against your intended stack.\n\n## Who should try it\nA good fit is a team with a document corpus and an existing model budget, willing to run Docker Compose or a Helm chart and choose its own LLM, embedding and vector backend. It suits internal Q&A plus light agent tasks, and wiki upkeep of shared knowledge. A poor fit is someone wanting a single-binary local tool with no infrastructure, or a pure Python project that wants to import a retrieval library into existing code. If you would rather not self-host, the [WeChat Dialog Open Platform](https:\u002F\u002Fchatbot.weixin.qq.com\u002Flogin) and Tencent Cloud Lighthouse templates are documented alternatives, though the former is aimed at WeChat scenarios.\n\n## Requirements and limitations\nThe standard path needs Docker, Docker Compose and Git, plus model and storage configuration; the [project instructions](https:\u002F\u002Fgithub.com\u002FTencent\u002FWeKnora) list the profiles and upgrade steps. Expect ongoing upkeep: model providers, vector database and object storage are all choices you own, and the upgrade process recreates containers by version, so pin `WEKNORA_VERSION` and read release notes. There is no README cost estimate, so budget infrastructure and model usage yourself.\n\nThe license badge says MIT while the repository metadata reports NOASSERTION, so confirm terms before adopting it in a commercial product. Feature depth varies by integration; the enormous backend list should be verified rather than assumed equivalent.\n\n## HotRepo's take\nHotRepo recorded 35,308 stars on 2026-10-24 (+98) and 35,312 on 2026-10-25 (+94), its only two daily-board observations; that history is too thin to judge whether interest persists. Before committing, verify that the specific vector database and LLM vendor combination you need appears in the supported-backends documentation.",{"slug":25,"name":26},"ai-engineering","AI Engineering",[28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47],"agent","agentic","ai","chatbot","dsh-plugin","embeddings","evaluation","generative-ai","golang","knowledge-base","llm","multi-tenant","ollama","openai","question-answering","rag","reranking","semantic-search","vector-search","wiki","NOASSERTION","https:\u002F\u002Fweknora.weixin.qq.com","2026-10-25T02:45:47Z",[52,56],{"date":53,"stars":54,"stars_gained":55},"2026-10-24",35308,98,{"date":57,"stars":11,"stars_gained":13},"2026-10-25",1792915094]