Embedded vector DBs in LLM apps: where ChromaDB, LanceDB, Qdrant Edge hit limits
InsideDebt6345 · reddit · 2026-09-09
A practitioner summarizes recurring practical limits of embedded vector stores in LLM apps across memory behavior, write-path correctness, and offline use:
- ChromaDB: HNSW lives in RAM; latency and correctness degrade at scale or with multiple writers.
- LanceDB: strong for multimodal and object storage, but concurrent writes conflict without serialization.
- Qdrant Edge: solid offline library with hybrid search; production concurrency limits are poorly documented, best when syncing to a central server.
Embedded stores look easy until agents write concurrently or hardware is constrained; the post outlines when to move from embedded to a separate process on air-gapped hardware. Disclosure: the author works with Actian on VectorAI DB, so treat it as informed but interested commentary.
More from Infra
- OpenAI plans ~$1T compute spend; expert sees it fueling AGI, not inference — NinaDSchick · 2026-09-09
- Compute markets split: financial players want standardized exposure, operators want bespoke delivery — AccBalanced · 2026-09-09
- Humanoid revolution needs $250B more motors on top of $50-60B in use today — stepjamUK · 2026-09-09
- Google Cloud breaks down 4 ways to serve open models, from managed to fully yours — rseroter · 2026-09-09
- Dell AI Factory vs cloud: agent workloads pay back in as little as 2.1 months, Signal65 finds — ryanshrout · 2026-09-09
- Spotify's Portal plugin claims to cut Claude Code token costs by 90% — fsharpman · 2026-09-09