An open-source context layer for building AI on top of company data

We’ve been building PipesHub for a while now, and I’d love to get more teams to try it and tell us where it breaks. The problem we kept running into was pretty simple: Building an AI app over company data looks easy in a demo. Connect a few sources, chunk the documents, throw them into a vector DB or connect to few mcps, add an LLM. Then you try to make it actually useful. You have data spread across S3, Google Drive, Slack, Jira, Confluence, SharePoint, email, databases, etc. Permissions need to be preserved. Documents change. The same file shows up in multiple places. Citations need to point back to the actual source. And eventually you want agents and other applications to use all of this context without rebuilding the same integration layer every time. That’s what we’re trying to solve with PipesHub with first class support open models. It’s an Apache 2.0 open-source context layer that connects to your company data and makes that context available to search, chat, agents, MCP clients, or your own applications. A few things we care about: Self-host it on your own infrastructure Preserve source permissions Get citations back to the original documents Combine knowledge graph + semantic retrieval Bring your own LLM and embedding models Use it from Python, TypeScript, Go, or MCP Avoid locking yourself into one database or infrastructure stack First class support for Open Models We also deliberately kept the core infrastructure pluggable: Layer Options Graph DB Neo4j, ArangoDB Vector DB Qdrant, OpenSearch, Redis Message broker Kafka, Redis Streams KV / config Redis, etcd Blob storage Local filesystem, S3, Azure Blob Models Your choice of LLM + embedding provider, including local models If you already have Qdrant and redis running, you can keep using them. Prefer Neo4j over ArangoDB? That's totally fine. Want to run the models locally? You can do that too. The goal is to give you one context layer without forcing you to adopt our entire stack. While building this, we’ve had to solve a bunch of problems that only become obvious once you move beyond simple agents: permission-aware retrieval, keeping citations accurate through the pipeline, deduplicating the same content across sources, efficiently re-indexing changed documents, making indexing behave well across very different workloads and more. Some of the solutions we ended up with are fairly unconventional, and I’d be happy to write more about them or discuss the trade-offs with anyone working on similar systems. There’s still plenty we want to improve, which is also why I’m posting this. If you’re building internal AI tools, enterprise search, RAG, workflow automation or agents that need access to company knowledge, I’d really appreciate it if you gave PipesHub a spin. GitHub: github.com/pipeshub-ai/pipeshub-ai Install: curl -fsSL get.pipeshub.com/install | bash If you try it and something feels unnecessarily complicated, slow, broken, or just badly designed, tell us.