kynetradb
One Rust binary: BM25 search + vector + KV + document + auth + files + realtime + agentic admin.
vs
Chroma
Open-source AI-native vector database built for LLM applications — simple Python-first API.
Feature comparison
| Dimension | kynetradb | Chroma |
|---|---|---|
| Full-text search | BM25 (parallel, 1.07 ms @ 100k) | None |
| Vector search | Brute-force cosine (2.21 ms @ 100k, no HNSW yet) Chroma uses HNSW which scales better past ~100k vectors | HNSW |
| Auth | Built-in — email+password, refresh tokens, OTP/magic-link, OAuth (Google/GitHub), anonymous, /auth/v1 | None |
| Row-Level Security | Built-in — auth.uid() DSL, per-row enforcement, SELECT/INSERT/UPDATE/DELETE | No |
| File storage | Built-in — buckets, public/private, signed URLs, object RLS, local + S3-compatible | None |
| Realtime | WebSocket — postgres_changes (INSERT/UPDATE/DELETE), presence, broadcast; RLS-filtered | None |
| Edge Functions | Built-in — WASM /functions/v1, sandboxed, 10s timeout, SDK .invoke() | No |
| TypeScript SDK | @kynetra/client — supabase-js-compatible, from().select().eq() | No official client SDK |
| KV lookups | Yes (point lookup by ID) | No |
| Document filter | Yes (JSON predicates) | Yes |
| LLM runtime | Yes (Anthropic + OpenAI + Ollama) | No |
| Outbound DB sync | Yes (12 sinks: Postgres, DynamoDB, BQ, Firestore, CF, Mongo, Redis, Pinecone) | No |
| Self-host | Yes (single binary) | Yes |
| Single binary | Yes | No |
| License | Apache-2.0 | Apache-2.0 |
| Deploy targets | 18 (1-click) | 0 (1-click) |
| Free tier | Yes — Apache-2.0, self-host free | yes — self-host free |
When to pick Chroma
The fastest path from LangChain prototype to working vector search. Great DX; not designed for multi-tenant production or non-Python stacks.
- You need HNSW at scale past ~100k vectors — kynetradb uses brute-force today.
- Your team is already invested in Chroma's SDK and ecosystem.
When to pick kynetradb
- You want Supabase's developer experience — the
@kynetra/clientTypeScript SDK, PostgREST REST API, GoTrue-shaped auth, RLS, realtime channels — running as a single binary on infra you own. - You need BM25 full-text + vector similarity search built in with no separate service and no CDC pipeline.
- You want to deploy to 18 targets (including 5 Indian providers) from one Dockerfile.
- You need outbound sync to 12 databases (Postgres, DynamoDB, BigQuery, Firestore, Cloudflare, MongoDB, Redis, Pinecone) with zero extra code.
- You want an agentic admin with 10 typed LLM-driven actions and a persisted audit trail.
- You want Apache-2.0 with a self-host path that doesn't require an ops team.
- You want a single binary with no runtime dependencies — no container fleet to operate.
Vector upsert + query — both APIs side by side
Upsert an embedding and query nearest neighbours. These are documentation-accurate shapes, not runnable end-to-end examples.
kynetradb
# kynetradb — upsert entity with embedding
curl -X POST https://your.host/v1/entities \
-H "Authorization: Bearer $KEY" \
-H "Content-Type: application/json" \
-d '{
"kind": "product",
"attrs": { "title": "Aurora Espresso" },
"embedding": [0.1, 0.2, 0.3, 0.4]
}'
# kynetradb — vector similarity query
curl -X POST https://your.host/v1/vector \
-H "Authorization: Bearer $KEY" \
-H "Content-Type: application/json" \
-d '{
"vector": [0.1, 0.2, 0.3, 0.4],
"top_k": 10
}' Chroma
# Chroma — Python client
import chromadb
client = chromadb.Client()
collection = client.get_or_create_collection("products")
# Upsert
collection.upsert(
ids=["aurora-espresso"],
embeddings=[[0.1, 0.2, 0.3, 0.4]],
metadatas=[{"title": "Aurora Espresso"}],
)
# Query
results = collection.query(
query_embeddings=[[0.1, 0.2, 0.3, 0.4]],
n_results=10,
)