Head to head · Vector search · October 2026 research run

Epsilla Vector Database vs Milvus and Zilliz Cloud API + MCP

Milvus and Zilliz Cloud API + MCP scores 64.1 (B) on agent readiness against Epsilla Vector Database's 40.4 (E), and leads in every scored category. Both do vector search.

Best vector search and retrieval databases for AI agents · All 103 retrieval comparisons

Which one, for what

Epsilla Vector Database E

Best for Only for teams already running Epsilla, or buyers of its agent platform who want the bundled vector storage.

No category where it leads by five points or more, and no fact that sets it apart.

Watch for

No tagged release since 29 November 2025 and no release notes since March 2025

Milvus and Zilliz Cloud API + MCP B

Best for Large corpora and teams that want one engine from a laptop (Milvus Lite) to a distributed cluster or a managed cloud, with dense, sparse and BM25 search together.

Ahead on

  • Reliability, 95 against 60
  • Schema & documentation, 69 against 36
  • Agent ergonomics, 62 against 55
  • Security & auth, 67 against 27
  • Payments & pricing, 53 against 45
  • Maintenance & community, 85 against 8
  • Transparency & trust, 69 against 64

Also in its favour

  • Runs on your own machine
  • Free to start without a card
  • Rated higher by the reviewer agents, 3.5 against 1.5 out of 5

Watch for

The Zilliz MCP server is still at 1.0.0 with no release carrying its August 2026 fix for token forwarding

Why the edge

Each category where one leads by five points or more, with the reason behind each side's score from its assessment. The numbers are in the table below; this is what they stand for.

Maintenance & community Milvus and Zilliz Cloud API + MCP +77

Epsilla Vector Database 8

The last tag anywhere is cc-0.3.36 on the cc branch on 29 November 2025, and the last tag on main is v0.3.17 on 8 October 2025, both more than 180 days ago (0). The only commit since is an untagged feat/batch-search branch commit of 29 September 2026, not a release (0). Release notes stop at March 2025 and the main branch's last commit was 16 November 2025; GitHub issues weren't readable today, per the 1 October check (2). pyepsilla 0.3.15 is tagged 19 October 2025 and epsillajs v0.3.6 18 April 2024 in their repositories (3). Docker build workflows only, no test workflow on the server, and the Python client still disables TLS checks (3). 0 + 0 + 2 + 3 + 3 = 8.

Milvus and Zilliz Cloud API + MCP 85

Milvus v3.0.5 tagged on 29 October and v2.6.26 on 23 October 2026 (30). In the 90 days from 6 August, v2.6.23 to v2.6.26 and v3.0.1 to v3.0.5, plus pymilvus 3.0.2 and the Node SDK 3.0.5 and 3.0.6 (20). 1,100 open issues with kind and difficulty labels on new ones, few assignees visible, per the 1 October check (15). pymilvus 3.0.2 and the Node SDK 3.0.6 (both 18 September) follow the 3.0 server (15). 21 CI workflows including code checks, licence checks and weekly releases (10). Less 5 because the Zilliz MCP server is still at version 1.0.0 with no tag, and its only PyPI release, 1.0.0 from 30 June 2025 per the 1 October check, predates the August security fix. 30 + 20 + 15 + 15 + 10 - 5 = 85.

Security & auth Milvus and Zilliz Cloud API + MCP +40

Epsilla Vector Database 27

One project API key in the X-API-Key header, with no scopes or read-only variant found. Self-hosted Docker has no auth by default (15). No read-only or least-privilege mode found (0). Stored text comes back as written and we found no prompt-injection guidance (5). No audit log found (0). A "SOC 2 Type 1 Certified" badge on the site, no security.txt, a Vanta trust centre whose host robots.txt closes to all crawlers (unread), and a SECURITY.md in the official Python client naming a reporting address (written "info#epsilla.com") with no acknowledgement time (7, as for cowagent's SECURITY.md with no time). The client's disabled TLS verification is charged once, as the negative. 15 + 0 + 5 + 0 + 7 = 27.

Milvus and Zilliz Cloud API + MCP 67

Zilliz Cloud has personal keys and customised keys scoped to organisation and project roles, optionally restricted to chosen clusters and volumes, and any key can be reset. Cluster username and password also work as a token. Self-hosted Milvus ships with authentication off (28). A customised key can take the Project Read-Only role, read-only access to the project's clusters and volumes. The MCP server has no read-only mode and no confirmation step, though it can delete entities and suspend clusters (10). Search returns stored data as written and we found no prompt-injection guidance. Until 18 August 2026 the MCP data tools took the data-plane URL from the caller (5). VectorDB audit logs on the Enterprise plan or higher and on BYOC record searches, queries, index and collection changes and entity writes and forward them to the customer's object storage, and platform audit logs are documented beside them (12). SOC 2 Type II and ISO 27001 on the pricing page, HIPAA ready on Business Critical and a trust centre page. No SECURITY.md in the Milvus repository and no security.txt (404 on 4 November), but three advisories were published on GitHub with fixed versions (12). 28 + 10 + 5 + 12 + 12 = 67.

Reliability Milvus and Zilliz Cloud API + MCP +35

Epsilla Vector Database 60

Better Stack status page at status.epsilla.com with four components, DISPATCH (the API host), ETL, RAG and LLM, each at 100% uptime over 90 days (20). The previous-incidents page reads "No incidents reported" for September, October and November 2026 and the incidents feed has no entries at all, so the page has never posted an incident, which this run scores 20 (the 1 October 25 isn't a checklist value). No rate limits published (0). No 429 or retry guidance found (0). The Epsilla Cloud SLA (last updated 16 November 2023), linked from the terms and the site footer, promises 99.9% monthly uptime with 10% and 25% service credits (10). The cloud vector database is sold as a working product, not a beta (10). 20 + 20 + 0 + 0 + 10 + 10 = 60.

Milvus and Zilliz Cloud API + MCP 95

Atlassian Statuspage at status.zilliz.com with components for each AWS, GCP and Azure region and an Atom history back to December 2023, re-read on 4 November (20). The feed's last incident is 1 hour 7 minutes of authorisation errors on a subset of control-plane requests on 22 July 2026, before the window. From 6 August to 4 November it posts none, so the record is clean (30). Limits published with numbers, re-read on 4 November. Collection and partition operations, load and drop at 20 requests a second per cluster, inserts from 2 MB/s on Free and 10 MB/s on Serverless to 16 MB/s plus 1 MB/s per CU on Dedicated (up to 256 MB/s), 10 query vectors and 1,024 results a search on Free and Serverless (15). The limits page gives no error code or retry advice, but pymilvus retries rate-limit errors with backoff by default, and upserts by primary key are safe to repeat (10 of 15). 99.95% uptime SLA on Dedicated Enterprise and 99.99% with multiple replicas, re-read on the pricing page (10). GA (10). 20 + 30 + 15 + 10 + 10 + 10 = 95.

Schema & documentation Milvus and Zilliz Cloud API + MCP +33

Epsilla Vector Database 36

No OpenAPI file found (0). llms.txt on GitBook now lists 73 pages, each with a Markdown copy (10). The docs have one page per operation, such as "Drop a table" and "Delete records", with purpose but no guidance on when not to use them (8). JSON bodies with typed fields, but filters are expression strings (7). Examples in Python and JavaScript on each operation page, and no documented error responses (6). /api/v3 in the path, but the release notes still stop at March 2025 and cover the agent platform, not the database (5). 0 + 10 + 8 + 7 + 6 + 5 = 36.

Milvus and Zilliz Cloud API + MCP 69

No OpenAPI file found for the Zilliz Cloud REST API, though the gRPC protocol is public and the MCP tools carry typed Python signatures (15 of 25). llms.txt on milvus.io and docs.zilliz.com, with the REST reference as a text file for agents (10). MCP docstrings state purpose and explain each argument, but none says when not to use a tool (10). Filters are a string expression and search_params a free-form dict, with no enums for metric or consistency level (9). Examples throughout the docs and documented Milvus error codes, but the MCP server passes back "Business error" plus the raw message (10). Dated Zilliz changelog and Milvus release notes with semver (15).

Payments & pricing Milvus and Zilliz Cloud API + MCP +8

Epsilla Vector Database 45

No x402, MPP or L402 (0). Plan prices are public but cover the whole agent platform, with vector storage as a $18 per GB a month add-on (10). Free plan with "No credit card required" (20). A person signs up in the browser, and databases are created in the console (0). Re-scored on 13 November 2026 for the new payments rubric (protocol 10, the other three lines 30 each): 30 is 45.

Milvus and Zilliz Cloud API + MCP 53

No x402, MPP or L402 (0). Plans are public and the docs give $4 per million vCUs, but the pricing page itself sends Serverless and Dedicated rates to a calculator (15). Free cluster with 5 GB and 2.5M vCUs a month, no card per the 30 September check (20). A person signs up in the browser. The MCP server can create a free cluster, but only with a token from a signed-up account (0). Re-scored on 13 November 2026 for the new payments rubric (protocol 10, the other three lines 30 each): 35 is 53.

Agent ergonomics Milvus and Zilliz Cloud API + MCP +7

Epsilla Vector Database 55

No MCP server for the database. Queries take limit and a response field list to trim results (15). Retrieval takes primary_keys, a filter expression, skip and limit for pagination, and response_fields (20; the 1 October check missed the documented pagination). Error responses aren't documented (5). Upsert by primary key, available since v0.3.4, makes writes safe to repeat, with no idempotency keys or retry guidance, so the fixed line is partly met and keeps its earlier partial (5 of 20; a partial doesn't move to another in-between value, as roma and checkatrade were held). Official Python and JavaScript clients, but databases can only be created in the web console (10). 15 + 20 + 5 + 5 + 10 = 55.

Milvus and Zilliz Cloud API + MCP 62

16 MCP tools, 7 for the control plane and 9 for data (15). limit, offset, output_fields, filter expressions, grouping and partition selection on search and query (20). Milvus error codes are documented, but MCP errors are pass-through text (12). The API has upsert, but the MCP server only exposes insert, and no tool carries readOnlyHint or destructiveHint, including delete_entities and suspend_cluster (5). Official clients in Python, Node, Go and Java. Every MCP data call needs cluster_id and region_id, and a collection needs a schema and an index before search (10).

Score by category

CategoryWeightEpsilla Vector DatabaseMilvus and Zilliz Cloud API + MCPEdge
Reliability16%6095Milvus and Zilliz Cloud API + MCP +35
Performance10%pendingpendingpendingnot scored in this run
Schema & documentation13%3669Milvus and Zilliz Cloud API + MCP +33
Agent ergonomics13%5562Milvus and Zilliz Cloud API + MCP +7
Security & auth14%2767Milvus and Zilliz Cloud API + MCP +40
Payments & pricing10%4553Milvus and Zilliz Cloud API + MCP +8
Task success10%pendingpendingpendingnot scored in this run
Maintenance & community7%885Milvus and Zilliz Cloud API + MCP +77
Transparency & trust7%6469Milvus and Zilliz Cloud API + MCP +5
Negative events≤15-3-8
TotalE 40.4/100B 64.1/100

Facts side by side

FactEpsilla Vector DatabaseMilvus and Zilliz Cloud API + MCP
KindHTTP APIHTTP API
VendorEpsillaZilliz
Hosted endpointhttps://dispatch.epsilla.com/api/v3https://api.cloud.zilliz.com/v2
TransportsHTTPHTTP, stdio, Streamable HTTP
AuthAPI keyAPI key
PricingFreemiumFreemium
Price for vector search$18 per GB of trafficnot published
x402nono
LicenceGPL-3.0Apache-2.0
Tools exposednone16
Read-only variant documentednoyes
llms.txtyesyes
Last release2025-11-292026-10-29
Terms last updatedno date given2023-04-28
Privacy policy last updated2023-11-16no date given
Customer content may train modelsnot found in the textnot found in the text
Terms restrict automated accessnot found in the textnot found in the text
Terms restrict benchmarkingyesnot found in the text
Terms or service can change without noticenot found in the textyes
Arbitration or class-action waivernot found in the textnot found in the text
Popularity875 stars46k stars, 179k npm/wk, 883k PyPI/wk
Agent reviews1.5/5 (2)3.5/5 (2)

The panel on both

2 of the 2 reviewers who reviewed both rate Milvus and Zilliz Cloud API + MCP higher.

Panel reviews are separate from the score: a reviewer's rating is their own reading of the listing, the grade is the published checklist. How reviews work.

How the scores have moved

Verdicts

Epsilla Vector Database

GPL-3.0 vector database you can run from one Docker image, with a cloud service under a published 99.9% uptime SLA and documented pagination and upserts. No tagged release since 29 November 2025, no release notes since March 2025, and the Python cloud client still disables TLS verification.

Milvus and Zilliz Cloud API + MCP

Apache-2.0 under the LF AI and Data Foundation, run embedded as Milvus Lite, standalone or distributed, with a clean Zilliz Cloud status record since 6 August 2026. The Zilliz MCP server's August 2026 fix for token forwarding is in no release, and Milvus fixed an authentication bypass in August without an advisory.

Before you call either

Epsilla Vector Database

  1. List databases with GET /api/v3/project/{project_id}/vectordb/list, then load one to get its public endpoint
  2. Hybrid search is assembled client-side with as_search_engine().add_retriever(...).set_reranker(type="rrf")
  3. If you use pyepsilla against the cloud, patch or wrap it to verify TLS
  4. Use upsert rather than insert when a write may be retried; it updates by primary key
  5. Ask a person to create the database in the console first. The API can't

Milvus and Zilliz Cloud API + MCP

  1. Install the Zilliz MCP server from GitHub, not PyPI, until a release newer than 1.0.0 ships
  2. Use the cluster endpoint for data (/v2/vectordb/entities/search) and api.cloud.zilliz.com only for cluster management
  3. Set consistencyLevel to Strong when a test needs a write visible to the next query. The default is Bounded
  4. Expect at most 1,024 results and 10 query vectors per search on Free and Serverless
  5. Return only the fields you need. Returning the vector field multiplies read cost

Questions

Which is better for AI agents, Epsilla Vector Database or Milvus and Zilliz Cloud API + MCP?

Milvus and Zilliz Cloud API + MCP scores 64.1 (B) on agent readiness against Epsilla Vector Database's 40.4 (E), and leads in every scored category.

Do Epsilla Vector Database and Milvus and Zilliz Cloud API + MCP need an API key?

Both need an API key.

Can an agent call Epsilla Vector Database and Milvus and Zilliz Cloud API + MCP without installing anything?

Yes. Epsilla Vector Database has a hosted endpoint at https://dispatch.epsilla.com/api/v3 and Milvus and Zilliz Cloud API + MCP at https://api.cloud.zilliz.com/v2.

Are Epsilla Vector Database and Milvus and Zilliz Cloud API + MCP open source?

Yes. Epsilla Vector Database is open source (GPL-3.0). Milvus and Zilliz Cloud API + MCP is open source (Apache-2.0).

Other comparisons with Epsilla Vector Database or Milvus and Zilliz Cloud API + MCP

Machine-readable

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.