SwarmV Python SDK

pip install "swarmv[fast]"   # [fast] adds orjson + numpy for max throughput

from swarmv import SwarmV

with SwarmV(api_key="...") as vec:
    vec.insert({"id": "1", "vector": embedding, "metadata": {"lang": "en"}})  # ids: numeric strings
    results = vec.search(query_embedding, top_k=5, filter={"lang": "en"})
    vec.delete("1")

Async, with concurrent batching:

from swarmv import AsyncSwarmV

async with AsyncSwarmV(api_key="...") as vec:
    batches = await vec.search_many(queries, top_k=5, concurrency=64)
    await vec.insert_many(records)

Feature parity with the TypeScript SDK: insert, search (top_k, nprobe, filter), delete, and SemanticCache (plus AsyncSemanticCache). Python-only extras: insert_many, search_many, numpy input.

Temporary failures (429, 5xx including "write backlog full" during heavy bulk loads, network errors) are retried automatically, waiting as long as the server's Retry-After asks: max_retries=6 by default. Indexes grow to 1,000,000 vectors (split into shards automatically past 100,000); see https://swarmv.dev/docs for limits and errors.

Performance: pooled HTTP/2 keep-alive connections, optional orjson, no per-element Python loops for numpy arrays. Reuse one client per process.