Milvus vs pgvector
Milvus vs pgvector: side-by-side comparison of two vector-database infrastructure systems — architecture, taxonomy, license, pricing, MCP/A2A support, and direct edges.
Cost & capability
| Milvus | pgvector | |
|---|---|---|
| Cost tier | free | free |
| $/Mtok input | 0 | 0 |
| $/Mtok output | 0 | 0 |
Where they differ (15)
Rows where both sides have data and the values disagree — the shortlist of dimensions that actually distinguish these two systems.
| Milvus | pgvector | |
|---|---|---|
| Type | Multi-vector + hybrid + GPU | PostgreSQL vector extension |
| Created | 2019-09 | 2021-04 |
| Latest release | v3.0.3 (2026-10-14) | v0.8.6 (2026-07-29) |
| License | Apache-2.0 | PostgreSQL |
| GitHub | milvus-io/milvus — ~46k stars, last commit 2026-10 | pgvector/pgvector — ~23.5k stars, last commit 2026-07 or later (0.8.7 unreleased) |
| Pricing | free (OSS); managed via Zilliz Cloud (free tier, then usage-based) | free (OSS) |
| Funding | $113M total (Zilliz: $43M 2020 + $60M 2022); Zilliz is the commercial entity | No external funding — open-source project by independent developer |
| Backend storage | custom (object-storage backed; etcd metadata; Pulsar/Kafka log) | Postgres |
| Deployment | Both (self-hosted OSS + Zilliz Cloud managed) | Self-hosted only — PostgreSQL extension; also offered inside managed Postgres (Supabase, AWS RDS, Azure, Neon, etc.) |
| API surface | REST, gRPC, SDK: Python, Java, Go, Node.js, C#, Ruby | SQL via Postgres protocol |
| Embedding | multiple supported | BYO |
| Multi-tenancy | Control plane and data plane in separate VPCs/subnets; BYOC option for hard tenant isolation | Logical isolation via Postgres schemas/databases; managed providers (Supabase, Neon, RDS) add tenant-level controls |
| MCP | yes — official zilliztech/mcp-server-milvus (GitHub) | yes — community Postgres MCP servers with pgvector tools (e.g. MOmayr/mcp-postgres-pgvector) |
| A2A | no Google A2A (Agent2Agent) integration documented as of 2026-05. | not supported |
| OpenTelemetry | first-class — Prometheus + OTel | via Postgres OTel |
At a glance
| Milvus | pgvector | |
|---|---|---|
| Section | Vector-database infrastructure | Vector-database infrastructure |
| Tier | T1 | T1 |
| Type | Multi-vector + hybrid + GPU | PostgreSQL vector extension |
| Created | 2019-09 | 2021-04 |
| Latest release | v3.0.3 (2026-10-14) | v0.8.6 (2026-07-29) |
| License | Apache-2.0 | PostgreSQL |
| GitHub | milvus-io/milvus — ~46k stars, last commit 2026-10 | pgvector/pgvector — ~23.5k stars, last commit 2026-07 or later (0.8.7 unreleased) |
| Pricing | free (OSS); managed via Zilliz Cloud (free tier, then usage-based) | free (OSS) |
| Funding | $113M total (Zilliz: $43M 2020 + $60M 2022); Zilliz is the commercial entity | No external funding — open-source project by independent developer |
| Backend storage | custom (object-storage backed; etcd metadata; Pulsar/Kafka log) | Postgres |
| Deployment | Both (self-hosted OSS + Zilliz Cloud managed) | Self-hosted only — PostgreSQL extension; also offered inside managed Postgres (Supabase, AWS RDS, Azure, Neon, etc.) |
| API surface | REST, gRPC, SDK: Python, Java, Go, Node.js, C#, Ruby | SQL via Postgres protocol |
| Embedding | multiple supported | BYO |
| Multi-tenancy | Control plane and data plane in separate VPCs/subnets; BYOC option for hard tenant isolation | Logical isolation via Postgres schemas/databases; managed providers (Supabase, Neon, RDS) add tenant-level controls |
| MCP | yes — official zilliztech/mcp-server-milvus (GitHub) | yes — community Postgres MCP servers with pgvector tools (e.g. MOmayr/mcp-postgres-pgvector) |
| A2A | no Google A2A (Agent2Agent) integration documented as of 2026-05. | not supported |
| OpenTelemetry | first-class — Prometheus + OTel | via Postgres OTel |
| Optimised for | low-latency similarity search + scale | low-latency similarity search + scale |
| Anti-fit | not for relational / graph-heavy queries (vector-first by design) | not for relational / graph-heavy queries (vector-first by design) |
Pros & cons
Milvus
Pros: Largest-scale OSS vector DB — billions of vectors comfortably; CNCF graduate; strong community in China + research labs.
Cons: Operational complexity is high vs Pinecone-managed; documentation can be uneven across language clients.
pgvector
Pros: Postgres extension — keeps vector data alongside transactional data with full ACID; minimal ops for shops already running Postgres.
Cons: Performance plateau at very large vector counts; limited filter pushdown vs purpose-built engines.