菜单
Storage backends and indexing for vector databases in Elixir. Use when choosing persistence, implementing exact vs approximate k-NN, or integrating pgvector, file, or ETS for vector storage.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
When publishing a new elix-db version or making version changes, run versioned sample use cases, collect memory/time/latency, compare to the previous version report, and update reports. Remember this workflow for releases.
Debug, verify, and compare elix-db to industry after each plan step. Use after implementing any plan step or changing vector/store/API logic; run tests, IEx checks, and document efficiency vs Qdrant/Milvus/pgvector.
Elixir patterns for vector storage and similarity search. Use when implementing GenServer-based vector store, ETS index, Nx/Scholar distance math, or supervision for a vector DB in Elixir.
Design and implement a vector database in Elixir. Use when building embedding storage, similarity search, k-NN retrieval, or when the user mentions vector DB, embeddings, or semantic search in Elixir.
| name | vector-db-storage |
| description | Storage backends and indexing for vector databases in Elixir. Use when choosing persistence, implementing exact vs approximate k-NN, or integrating pgvector, file, or ETS for vector storage. |
| Backend | Use case | Persistence | Scale |
|---|---|---|---|
| ETS | Prototype, small index, low latency | No | < ~100k–500k vectors in memory |
| File | Simple persistence, no DB | Yes (binary/term) | Same as ETS, plus disk |
| PostgreSQL + pgvector | Production, SQL + vectors | Yes | Large; use IVFFlat/HNSW indexes |
| Qdrant (client) | Dedicated vector engine | Yes (external) | Very large |
{id, vector, metadata}; vector as list or binary.terminate/2 or :gen_server.cast(pid, :persist) → write ETS contents to file (e.g. :erlang.term_to_binary(entries) or custom binary format).Use pgvector and Ecto for persistent, indexed storage.
id, embedding (pgvector type), metadata (jsonb/map).embedding for approximate k-NN.<=>, <->) in raw SQL or Ecto fragments; limit k.# Example: nearest neighbors (cosine distance)
# fragment("embedding <=> ?::vector", ^Nx.to_flat_list(query))
| Method | Accuracy | Speed | When |
|---|---|---|---|
| Exact (scan all) | Exact | O(n) per query | Small n (e.g. < 100k), correctness critical |
| IVFFlat (pgvector) | Approximate | Fast | Medium/large, good recall with tuning |
| HNSW (pgvector) | Approximate | Very fast | Large, high recall |
For in-memory ETS, start with exact k-NN; move to pgvector (or Qdrant) when n grows or persistence is required.