My team's been using Searchable for about a year to index Confluence docs, Slack threads, and GitHub issues. It's... fine. The latency on queries has gotten worse as we've scaled, and the per-user pricing is starting to bite. With the recent surge in open-source embedding models and RAG frameworks, I'm convinced we can build or swap to something more controllable and cost-effective.
I've been benchmarking a simple stack locally:
* **ChromaDB** or **Qdrant** for the vector store, running in our k8s cluster.
* **sentence-transformers/all-MiniLM-L6-v2** for embeddings (good balance of speed/accuracy for our doc size).
* **LlamaIndex** to handle the ingestion pipeline and query construction.
A basic proof-of-concept ingestion script looks like this:
```python
from llama_index import VectorStoreIndex, SimpleDirectoryReader
from llama_index.vector_stores.qdrant import QdrantVectorStore
import qdrant_client
client = qdrant_client.QdrantClient(path="./qdrant_data")
vector_store = QdrantVectorStore(client=client, collection_name="docs")
documents = SimpleDirectoryReader("./company_docs").load_data()
index = VectorStoreIndex.from_documents(documents, vector_store=vector_store)
```
The query performance is on par with Searchable for our internal jargon, and we own the data pipeline. The real cost is the engineering time to maintain it.
What I need to figure out is the production-ready orchestration and access control layer. So my question to the community: what are you actually running in production for this use case?
* Are you using a managed service like Pinecone or Weaviate, or self-hosting?
* How are you handling document updates and incremental indexing?
* What's your solution for user authentication and row-level security (e.g., ensuring engineers only see docs for projects they have access to)?
I'm leaning towards a self-hosted Qdrant with a lightweight FastAPI wrapper to handle auth, but I don't want to rebuild a full SaaS platform. If there's a solid open-source project that bundles this all together, I'd rather use that.
-shift
shift left or go home