Zilliz Cloud

PARTIAL consensus from 3 PARTIAL.

The quickstart covers connection through search but assumes prepared external data instead of showing a direct vector insert. Cluster creation is fully documented on /docs/create-cluster with both console and REST paths, but the page titled Quickstart to Serving Cluster begins after cluster creation, never shows insert() at all (its only data step is a bulk import from an S3 bucket with access keys), and hardcodes the Dedicated-only endpoint form that /docs/access-connection-endpoints says Free and Serverless clusters must not use. The Quickstart, the most obvious path for a first vector search, skips cluster creation and imports data from external S3 storage rather than inserting individual vectors, forcing an agent to triangulate across three pages with no unified from-zero walkthrough.

PARTIAL consensus from 3 PARTIAL.

The docs give collection, partition, and insert limits but omit Free DDL and conflict on load rates; /docs/limits-and-restrictions did not load. One canonical page gives collection limits per tier (Free 5, Serverless 100, Dedicated 1,024 per CU capped at 16,384), a tier-independent per-collection partition cap of 1,024, and insert rates per tier (Free 2 MB/s, Serverless 10 MB/s, Dedicated 16 MB/s + 1 MB/s per CU), with no contradictions against the FAQ or free-trial pages, but the 20 req/s DDL rate limit is scoped to Serverless and Dedicated only and the cluster-wide partition ceiling is a per-CU formula that cannot be evaluated for Free or Serverless because the same page states that the concept of CU only applies to Dedicated clusters. Cloud limits page covers all plans but the BYOC limits page omits Free and Serverless tiers entirely (no collection caps, no vCU limits, no capacity limits, no nq=10 cap), creating a contradiction that forces an agent to determine which page applies.

PARTIAL consensus from 1 PASS, 2 PARTIAL.

The error table gives both corrections, and API Keys explains scope changes; /docs/authentication did not load. Recovery from 96008 is exact and self-contained and matches the List Clusters (V2) spec default of 10 with minimum 1 and maximum 100, but 96016 only says to check the API key scope and no page maps a REST operation to the role it requires (the Authorization header is documented merely as an API key with appropriate privileges), so the fix requires jumping unlinked to /docs/manage-api-keys in Cloud Guides; separately, 185 of the 187 page URLs published inside https://docs.zilliz.com/llms/restful.txt carry a doubled path segment and return 404. Error codes are listed with their messages but recovery for 96016 stops at 'check your Apikey scope' with no link to API-key management docs or guidance on how to set the correct scope, requiring guesswork.

PARTIAL consensus from 1 PASS, 2 PARTIAL.

The guide covers webhook setup, shows a JSON notification body, and uses Send Test Message to verify configuration. Console configuration steps and an example alert payload with orgId, projectId, summary, level, and timestamp are documented, and the Create Alert Rule (V2) REST spec exposes a WEBHOOK action type, but the docs never state the HTTP method, content type, expected response code, retry behaviour, or the shape of the test-message payload, the only verification step is clicking an icon identified solely by a screenshot, and the notification-channel guide never mentions the REST alert-rule API whose webhook field is named url while every other channel uses webhookUrl. Webhook setup and test message are documented but the test-message payload format is not shown, no expected success/failure feedback is described, and the example payload is a real alert rather than a connectivity-test payload.

PARTIAL consensus from 1 PASS, 2 PARTIAL.

The flow is complete, but PyMilvus guidance conflicts between 2.6.9 and 2.6.12; /reference/python/MilvusClient did not load. Installation, the token format of an API key or colon-joined username and password, and create_collection plus list_collections are all documented and mutually consistent, but every runnable Python example sets uri to a host that matches none of the three endpoint patterns printed on the same page, and the list_collections page opens by saying the method applies to dedicated serving clusters and on-demand compute while immediately listing Free and Serverless endpoints for it. PyMilvus installation, MilvusClient initialization with cluster credentials, and collection creation are documented with consistent, copyable code examples across two pages with no contradictions.