Weaviate Problems & Issues
SaasThe AI-native database for a new generation of software
Should you use Weaviate?
TryWeaviate is worth trying for most users (score 72/100). Most reported issues: Performance degrades at scale (18M+ objects); Memory usage grows heavily with data; Data corruption rendered database unusable. Not a good fit for: Teams needing easy scaling without DevOps. Users most often switch to Elasticsearch, Milvus, pgvector.
Top problems
- Performance degrades at scale (18M+ objects)
- Memory usage grows heavily with data
- Data corruption rendered database unusable
Good for
- Teams needing hybrid search
- Companies wanting open-source flexibility
- Multi-modal AI applications
Avoid if you are
- Teams needing easy scaling without DevOps
Common alternatives: Elasticsearch, Milvus, pgvector
Overview
Weaviate is an open-source AI-native vector database with hybrid search capabilities. Supports both self-hosted and managed cloud deployments. Features GraphQL API, built-in ML modules, and multi-modal data support. Known for flexibility but criticized for complexity at scale.
Signals & Issues
Patterns extracted from real user feedback — not raw reviews.
GitHub issue #4572 documents scaling problems: 'With 18 million objects ingested, every new query was taking anywhere between 5 to 15 seconds per query.' The main complaint is 'performance when trying to scale things up' - feels fine for small datasets but latency becomes unpredictable at scale.
Users report 'memory usage keeps increasing heavily as more data is added to the cluster.' Some users report memory reaching 86.4% during indexing of large knowledge bases. Multi-node Kubernetes deployments particularly affected.
Users report 'recurring DEADLINE_EXCEEDED errors during batch imports.' gRPC timeouts frustrate users trying to ingest large datasets. Requires tuning and workarounds that aren't well documented.
G2 reviews note Weaviate's 'resource-intensive nature' and 'the need for dedicated resources.' Users report difficulty getting it to 'run and scale reliably, potentially requiring considerable time and engineering effort to make Weaviate-based applications production-ready.'
G2 reviews suggest 'improving documentation as it's crucial for integrating and using the services easily.' Gaps exist 'particularly for AI docs for direct API access.' New users struggle with advanced configurations.
Hacker News users complain about 'needing to reindex all data when wanting to attach a new module (vectorizer, generator, etc) to a data class, which cramps your style when talking millions to billions of docs.' Module changes are disruptive.
Users describe the management console as 'incredibly barebones, with no way to turn off instances and limited ability to self-help when issues occur.' The web interface is not great, though client libraries work well.
One G2 user reported 'data corrupted and Weaviate became entirely useless, retrieving different data at each request.' During this incident, support 'took a day to respond each time.' Serious reliability concern for production use.
G2 reviews cite 'integration challenges' particularly when 'incorporating Weaviate into existing systems not originally designed for semantic search or knowledge graphs.' RBAC integration with directory services also problematic.
G2 notes Weaviate's 'resource-intensive nature that can lead to increased infrastructure costs depending on dataset size and query complexity.' Self-hosted requires significant compute. Cloud pricing changed in Oct 2025, adding complexity.
What Users Love
Excellent hybrid search combining vectors and BM25
Weaviate excels at hybrid search with 'parallel execution model where vector and BM25 searches run simultaneously.' Uniquely offers relativeScoreFusion to retain nuances of original search metrics. Best-in-class for combining keyword and semantic search.
Quick to get started with minimal config
Product Hunt reviews praise how 'quick it is to get started with Weaviate without spending days on configs or setups.' Clear documentation with active community. Can prototype fast before hitting scale challenges.
Exceptional customer support from the team
Multiple testimonials highlight 'impeccable' customer support. Makers of Cortex praise 'hands-on, architecture-level support.' Zefi.ai credits 'early, sustained support.' Support quality is a consistent differentiator.
Free self-hosted option with BSD-3 license
Weaviate is open-source with BSD-3 license - commercial friendly. Can self-host for free. No vendor lock-in if you run your own infrastructure. Strong alternative to Pinecone's cloud-only model.
Multi-modal data support (text, images, video)
Supports vectors from multiple modalities. Users praise 'seamless semantic search across media' including video embeddings. Built-in modules for various vectorizers make multi-modal easy to implement.
GraphQL API with flexible querying
Weaviate's GraphQL API allows complex queries with filters, aggregations, and cross-references. More expressive than typical REST APIs. Good for applications needing sophisticated query patterns.
Pricing Plans
Users: Unlimited
Storage: Limited by your infrastructure
Limitations: No managed operations, manual scaling, no enterprise support, requires DevOps expertise
Users: Unlimited
Storage: Pay per vector dimension
Limitations: Shared infrastructure, limited customization vs Dedicated
Users: Unlimited
Storage: Committed capacity
Limitations: Locked into annual contract
Users: Unlimited
Storage: Dedicated resources
Limitations: Higher minimum spend, longer provisioning time
Features
core
- Hybrid Search (Vector + BM25)
- Multi-Modal Support
- Cross-References
integration
- GraphQL API
- Kubernetes Native
ai
- Built-in Vectorizers
security
- Multi-Tenancy
- RBAC
reliability
- Backups & Replication
performance
- Product Quantization
Who Should Use This?
Good Fit
Teams needing hybrid search
Weaviate's parallel vector + BM25 search is best-in-class. relativeScoreFusion maintains nuances. If your use case requires combining keyword and semantic search, Weaviate excels.
Companies wanting open-source flexibility
BSD-3 license allows commercial use with modifications. Can self-host without vendor lock-in. Alternative to Pinecone's cloud-only model. Good for teams with DevOps capacity.
Multi-modal AI applications
Native support for text, image, and video embeddings. Built-in modules for various vectorizers. Great for applications combining different data types in semantic search.
Enterprises requiring data sovereignty
Self-hosted option keeps data in your infrastructure. Dedicated Cloud offers isolated environments. Good for regulated industries needing on-premise or specific region deployment.
Budget-conscious startups
Self-hosted is free (BSD-3). Cloud pricing at $0.10/hour is competitive (vs Pinecone $0.15, Qdrant $0.20). Can start cheap and scale up. Watch infrastructure costs if self-hosting.
Neutral
Small teams without vector DB experience
Quick to start prototyping, but steep learning curve for advanced usage. Documentation has gaps. Support is excellent if on paid plan. Start simple, but plan for complexity.
High-frequency update workloads
Module changes require full reindex which is disruptive. gRPC timeouts reported during batch imports. Works for moderate update frequency but test thoroughly for high-frequency scenarios.
Avoid
Teams needing easy scaling without DevOps
Scaling issues are well-documented - queries take 5-15 seconds at 18M+ objects. Memory grows heavily with data. Requires significant engineering effort for production-ready at scale. Consider Pinecone for managed simplicity.
Regret Patterns
Common buyer's remorse scenarios reported by users.
Worked great at 1M objects. At 18M+, queries took 5-15 seconds. Memory grew beyond provisioned capacity. Should have stress-tested with realistic data volumes before committing. Now migrating to Qdrant.
Chose self-hosted to save money. Spent weeks tuning Kubernetes deployment. Memory issues, gRPC timeouts, scaling challenges. Engineering time cost more than managed cloud would have. Should have started with Weaviate Cloud.
Needed to add a new vectorizer to existing data class. Discovered this requires reindexing all data. With millions of documents, this took days and required downtime. Should have planned module architecture upfront.
Database corrupted and returned different results each query. Support took a day per response during the crisis. Lost trust in reliability. Should have had better backup strategy and tested disaster recovery.
Assumed Weaviate would slot into existing architecture. RBAC integration with directory service was problematic. GraphQL learning curve for team used to REST. Should have evaluated integration requirements more carefully.
When It Breaks Down
Scenarios where this product tends to fail users.
Query latency degrades from milliseconds to 5-15 seconds. Memory usage becomes unpredictable. The 'main complaint is performance when trying to scale things up.' Requires significant infrastructure investment or migration.
Adding a module to existing data class requires full reindex. With millions of documents, this can take days. Causes service disruption. Better to plan module architecture from the start.
Memory tuning, Kubernetes configuration, and scaling become overwhelming. gRPC timeouts during imports. Time spent on operations exceeds saved costs. Consider Weaviate Cloud or simpler alternatives.
gRPC DEADLINE_EXCEEDED errors during large imports. System becomes unstable under write pressure. Need to implement batching strategies and backoff. Test import patterns thoroughly before production.
Management console is 'incredibly barebones' - can't turn off instances, limited self-service. When issues occur, you're dependent on support. Power users frustrated by lack of control.
Alternatives Users Switch To
Qdrant
8x mentionedCompanies switch for simpler scaling and Rust performance. Cognigy.AI switched to Qdrant for 'improved Knowledge AI experience.' Gain: better performance at scale, simpler operations. Trade-off: less mature hybrid search, smaller ecosystem.
Pinecone
7x mentionedTeams switch for fully managed simplicity. Gain: zero ops burden, easier scaling, sub-10ms latency. Trade-off: cloud-only (no self-host), vendor lock-in, higher costs at scale.
pgvector
6x mentionedPostgres teams switch to stay in existing stack. Gain: free, no new infrastructure, simpler architecture. Trade-off: less sophisticated hybrid search, fewer built-in AI features.
Milvus
5x mentionedEnterprises switch for massive scale. Gain: 'lightning-fast searches' at billion-vector scale, Apache 2.0 license. Trade-off: complex to operate, steeper learning curve than Weaviate.
Elasticsearch
4x mentionedTeams switch for mature ecosystem and broad functionality. Gain: battle-tested at scale, rich plugins, full-text + vector. Trade-off: complex operations, license concerns (SSPL).
Frequently Asked Questions
See how Weaviate compares in our Best Search Software rankings, or calculate costs with our Budget Calculator.
Work at Weaviate or spotted something outdated? Suggest a correction — we verify every change against public sources. See our methodology.