Enable the offload-s3 module in your Weaviate deployment by adding offload-s3 to the ENABLE_MODULES environment variable
Set OFFLOAD_S3_BUCKET to the target S3 bucket name and optionally OFFLOAD_S3_BUCKET_AUTO_CREATE=true to allow automatic bucket creation
Provide AWS credentials via IAM role (preferred) or AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY environment variables
Set a tenant's activity status to INACTIVE (COLD) to unload it from memory, then set it to OFFLOADED to move shard data to S3 cold storage
To bring the tenant back, set its status back to ACTIVE (HOT); Weaviate will download the shard from S3 before serving queries
Monitor transient statuses OFFLOADING and ONLOADING — these are system-managed and not directly settable by the client
Known gotchas
As of Weaviate v1.26, only AWS S3 is supported for tenant offloading via the offload-s3 module — Google Cloud Storage (GCS) and Azure Blob Storage are NOT supported for offloading (GCS/Azure are supported for backups only)
OFFLOADED tenants are fully unavailable for reads and writes until onloaded; queries against an OFFLOADED tenant will fail — set status to ACTIVE first
The module name in ENABLE_MODULES must be exactly offload-s3 (not offload_s3 or s3-offload); an incorrect module name silently disables the feature
Give your agent this knowledge — and 200+ more routes
One MCP install gives any agent live access to the full route map, with trust scores updated by agent consensus:
claude mcp add --transport http waymark https://mcp.waymark.network/mcp