A RAG pipeline that ends with posting to 17 social platforms
BullFlow turns your company docs into knowledge-grounded marketing content, then publishes it everywhere — with billing and multi-tenancy already built in.

What it does
推量牛 (BullFlow) is an open-source, MIT-licensed enterprise RAG system that runs the full chain: ingest company documents into a knowledge base, have a LangGraph ReAct agent write articles / image posts / short videos grounded in that knowledge, then one-click publish to 17 platforms — Douyin, Xiaohongshu, WeChat, Bilibili, plus LinkedIn, YouTube, X, Reddit and more. It ships as three systems: a user-facing app, an admin backend (23 menus), and an independently deployable publishing gateway.
The interesting bit
The entire RAG chain — OCR (PaddleOCR), semantic splitting (wtpsplit), embeddings (BGE-M3), reranking (BGE-Reranker), image search (OpenCLIP), transcription (WhisperX) — runs on local open-source models, so indexing and retrieval cost nothing and data never leaves your network. Only the final generation step calls an LLM, via an OpenAI-compatible endpoint you pick. The retrieval side is genuinely industrial: 12 input formats, a four-level parsing fallback chain, Milvus + Elasticsearch hybrid search with RRF fusion, and Ragas-based evaluation with automatic weight tuning.
Key highlights
- Answers cite retrieved chunks with relevance scores, traceable to source documents
- Content adapts to platform shape: mixed image-text articles for WeChat, 1–9 image posts for Xiaohongshu
- Scheduled rules generate and publish content unattended, with notifications on success or failure
- Full commercial layer: multi-tenant isolation, credit billing by token/image/second, subscriptions, payments, invoices, referral commissions
- Account safety features: sticky regional proxy IPs, max 15 accounts per IP per platform, health checks every 20 minutes
Caveats
- The README mentions a roadmap and “known limitations” section, but it’s cut off in the provided text, so current rough edges aren’t visible here
- Heavy operational footprint: MySQL, Milvus, Elasticsearch, Redis, and a separate gateway — this is a platform, not a script
- Publishing to Chinese platforms presumably rides on the companion gateway’s terms-of-service gray zones; the README doesn’t discuss that risk
Verdict
Worth a look if you’re building a content-marketing SaaS or running multi-account social matrices and want the boring parts (billing, tenancy, publishing) done for you. If you just want a clean RAG library to embed in your own app, this is far more machine than you need.
Frequently asked
- What is jaikydota/BullFlow?
- BullFlow turns your company docs into knowledge-grounded marketing content, then publishes it everywhere — with billing and multi-tenancy already built in.
- Is BullFlow open source?
- Yes — jaikydota/BullFlow is open source, released under the MIT license.
- What language is BullFlow written in?
- jaikydota/BullFlow is primarily written in TypeScript.
- How popular is BullFlow?
- jaikydota/BullFlow has 624 stars on GitHub.
- Where can I find BullFlow?
- jaikydota/BullFlow is on GitHub at https://github.com/jaikydota/BullFlow.