Weaviate v1.35.15: Production-Grade Replication and Backup Scaling

industry sources tracks production vector database releases closely because replication and backup reliability directly impact your application uptime. Weaviate v1.35.15 addresses three operational pain points that teams face at scale: async replication encoding, backup chunk splitting, and multimodal embedding support.

The async replication binary encoding improvements streamline how data replicates across cluster nodes without blocking writes. This reduces latency spikes during heavy replication loads - critical for builders running distributed deployments. The backup chunk size splitting enhancement lets you handle larger datasets without hitting memory or timeout issues during backup operations.

Gemini Embedding 2 Multimodal model support in the multi2vec-google module means you can now index and search across text and images using Google's latest embedding model directly within Weaviate. This opens workflows for visual search, cross-modal retrieval, and hybrid search applications that previously required external preprocessing.

• Async replication binary encoding: faster cross-node data synchronization without write blocking
• Backup chunk size splitting: handle larger datasets in backup operations more reliably
• Gemini 2 Multimodal in multi2vec-google: native text-image embedding without external conversion

For builders running Weaviate in production clusters, replication is non-negotiable - it's how you avoid data loss and ensure failover works. The async binary encoding improvement means your replication can keep pace with write throughput without creating bottlenecks. If you've hit replication lag under load, this release directly addresses that problem.

Backup chunk splitting matters if you manage vector databases larger than a few hundred million embeddings. Previous chunk size limits could cause backups to fail or timeout on large collections. This release lets you back up bigger datasets within your infrastructure constraints - essential for disaster recovery planning and compliance requirements.

If you're currently running v1.35.x, these improvements are bug fixes and optimization patches rather than breaking changes. Most teams should upgrade without friction. For users on older versions considering an upgrade, this release makes a stronger case for moving to the latest minor version.

• Replication lag reduction under write-heavy workloads
• Backup reliability for collections exceeding previous size limits
• No breaking changes - safe upgrade path for most deployments

Gemini 2 Multimodal support changes what you can build without external preprocessing. Before, searching across text and images required either separate embedding models or external APIs to convert images to embeddings. Now that's native within Weaviate using Google's latest multimodal model.

This matters for e-commerce builders creating visual search (find similar products from a photo), content platforms needing cross-modal discovery (find articles related to an image), and any application combining text and image data. The multi2vec-google module handles both modalities in a single embedding space, eliminating synchronization headaches.

Setup is straightforward if you're already using multi2vec-google - you configure the new model and reindex. If you're on other embedding providers, this gives you a reason to evaluate Weaviate's Google integration. Performance depends on your image resolution and text length, but native multimodal handling typically beats external pipeline approaches by 20-40% in latency.

• Text-image search in one embedding space without external conversion
• Native support eliminates pipeline complexity for multimodal applications
• Lower latency and simpler deployment compared to external embedding services

If you're running Weaviate in production, schedule an upgrade to v1.35.15 within your next maintenance window. The replication and backup improvements reduce operational risk - they're worth the testing cycle. Check your cluster replication metrics first, then run this in staging to verify backup performance on your actual dataset sizes.

For teams evaluating vector databases, this release reinforces Weaviate's focus on production operations. The improvements show they're listening to scaling pain points rather than just adding flashy features. If replication reliability or backup handling were concerns in your evaluation, v1.35.15 addresses those directly.

For builders considering multimodal search, test the Gemini 2 integration in a sandbox. Benchmark against your current approach - if you're using external embedding services or two-model pipelines, the latency and complexity reduction might justify switching. Document your results and share them with your team because multimodal is becoming table stakes for search UX.

The momentum in this space continues to accelerate.

• Schedule upgrade to v1.35.15 in next maintenance window - replication and backup improvements reduce operational risk
• Test Gemini 2 multimodal on your dataset to quantify latency and complexity gains
• Evaluate whether native multimodal support justifies migration from external embedding pipelines