An S3-native vector search engine. Object storage is the source of truth. Nodes are stateless.
- S3-native -- Object storage is the single source of truth
- Stateless nodes -- Any node can serve any query
- IVF indexing -- IVF-Flat, IVF-SQ8 (4x compression), IVF-PQ (16-32x), and Hierarchical ANN
- BM25 full-text search -- Inverted indexes with configurable tokenization, stemming, and multi-field
rank_byexpressions (opt-in, see below) - Bitmap pre-filters -- RoaringBitmap indexes for sub-millisecond attribute filtering
- Write-ahead log -- Durable writes with compaction into indexed segments
- Strong & eventual consistency -- Choose per-query (see Consistency semantics)
- Object storage -- S3, MinIO, and S3-compatible backends. GCS and Azure planned
Spin up Zeppelin with MinIO locally using Docker Compose:
docker compose upThis starts Zeppelin on port 8080 and MinIO on port 9000 with a pre-created zeppelin bucket.
# Server generates a UUID name — save it!
curl -s http://localhost:8080/v1/namespaces \
-H "Content-Type: application/json" \
-d '{"dimensions": 384}' | jq
# Response: {"name": "550e8400-e29b-41d4-a716-446655440000", "dimensions": 384, ..., "warning": "Save this namespace name..."}# Replace <ns> with the UUID from the create response
curl -s http://localhost:8080/v1/namespaces/<ns>/vectors \
-H "Content-Type: application/json" \
-d '{
"vectors": [
{"id": "vec-1", "values": [0.1, 0.2, 0.3, "...384 dims..."]},
{"id": "vec-2", "values": [0.4, 0.5, 0.6, "...384 dims..."]}
]
}' | jqcurl -s http://localhost:8080/v1/namespaces/<ns>/query \
-H "Content-Type: application/json" \
-d '{
"vector": [0.1, 0.2, 0.3, "...384 dims..."],
"top_k": 10
}' | jqcurl -s -X DELETE http://localhost:8080/v1/namespaces/<ns>/vectors \
-H "Content-Type: application/json" \
-d '{"ids": ["vec-1"]}' | jq| Method | Path | Description |
|---|---|---|
GET |
/healthz |
Liveness probe |
GET |
/readyz |
Readiness probe |
GET |
/metrics |
Prometheus metrics |
POST |
/v1/namespaces |
Create a namespace (returns UUID) |
GET |
/v1/namespaces/:ns |
Get namespace metadata |
DELETE |
/v1/namespaces/:ns |
Delete a namespace |
POST |
/v1/namespaces/:ns/vectors |
Upsert vectors |
DELETE |
/v1/namespaces/:ns/vectors |
Delete vectors |
POST |
/v1/namespaces/:ns/query |
Query nearest neighbors |
Use the HTTP API directly or generate a client from the canonical OpenAPI 3.1 spec. This repository does not maintain generated client SDK packages.
- Rust 1.85+
- Docker (for MinIO in tests)
cargo buildRun the default in-memory test suite:
cargo testFor the MinIO-backed integration pass:
docker compose -f docker-compose.test.yml up -d
TEST_BACKEND=minio \
TEST_S3_BUCKET=stormcrow-test \
MINIO_ENDPOINT=http://localhost:9000 \
MINIO_ACCESS_KEY=minioadmin \
MINIO_SECRET_KEY=minioadmin \
cargo test --tests# Start MinIO
docker compose -f docker-compose.test.yml up -d
# Run Zeppelin against local MinIO
STORAGE_BACKEND=s3 \
S3_BUCKET=stormcrow-test \
S3_ENDPOINT=http://localhost:9000 \
AWS_ACCESS_KEY_ID=minioadmin \
AWS_SECRET_ACCESS_KEY=minioadmin \
AWS_REGION=us-east-1 \
S3_ALLOW_HTTP=true \
cargo runcargo fmt --all -- --check
cargo clippy -- -D warningssrc/
storage/ Object store abstraction (S3, S3-compatible)
wal/ Write-ahead log: fragments, manifest, reader/writer
namespace/ Namespace CRUD and metadata
index/ Vector indexing (IVF-Flat with k-means)
cache/ Local disk cache with LRU eviction
compaction/ Background WAL-to-segment compaction
server/ Axum HTTP handlers, routes, middleware
Writes land in the WAL as immutable fragments. Background compaction merges fragments into indexed segments (IVF, bitmap pre-filters, BM25 inverted indexes). Queries probe the closest centroids and merge results from any un-compacted WAL fragments.
Consistency is selected per-query via the consistency field:
strongscans un-compacted WAL fragments in addition to indexed segments, so it always reflects writes committed to S3 — with one caveat: each node caches the namespace manifest for up to 500 ms. A write through node A is immediately visible to strong queries on node A (write-through cache), but a strong query on node B may miss it for up to the cache TTL. In other words: same-node read-your-writes; cross-node bounded staleness (≤ 500 ms). Single-node deployments get true strong consistency. If you need cross-node read-your-writes, sticky-route each client to one node.eventualreads indexed segments and applies delete tombstones from un-compacted WAL fragments, but skips WAL vector/BM25 scoring. Deletes are hidden immediately on the same node after the delete returns. Recent upserts and updates can still be stale until the next compaction cycle, so use it when write freshness within one compaction interval does not matter.
Writes are safe under concurrency without any coordinator: every manifest
commit is an ETag-guarded compare-and-swap, so concurrent upserts from
multiple nodes serialize correctly. Background compaction additionally
acquires a per-namespace lease (compaction.lease_duration_secs, default
300 s) so only one node compacts a namespace at a time; a fencing token +
CAS make even an expired-lease holder unable to commit stale state.
BM25 rank_by queries work out of the box against un-compacted WAL data.
For segment data, per-cluster and global inverted indexes are built during
compaction only when indexing.fts_index = true (or ZEPPELIN_FTS_INDEX=true)
— it is off by default because it adds compaction cost for namespaces
that never use FTS. Without it, segment BM25 falls back to a full scan,
which is rejected above bm25_max_full_scan_clusters (default 500) to
protect latency. Enable fts_index if you use rank_by at scale.
Licensed under the Apache License, Version 2.0.
Copyright 2026 Anup Ghatage. See NOTICE — redistributions and derivative works must retain the attribution notices it contains, per Section 4(d) of the license.