seekdb-js v1.2.0

Release date: 2026-03-26
Version: V1.2.0

New Features

• Embedded Mode: Added support for embedded mode.
• Sparse Vector Indexing: Added support for sparse vector indexing to improve keyword-based retrieval.
• Enhanced Index Configuration: Added more HNSW index configuration options and full-text search configuration options.
• ORM Adapters: Added support for Prisma and Drizzle adapters, extending SeekDB query and integration capabilities in embedded mode.

What's Changed

• feat: Integrate libseekdb to support embedded mode by @dengfuping in #7
• feat: add schema API and support Sparse embedding function by @raindrop93 in #22
• feat(seekdb): add client.execute API for raw SQL execution by @dengfuping in #20
• fix server mode batch operation by @raindrop93 in #30
• feat: add @seekdb/prisma-adapter and ORM examples (seekdb-drizzle, seekdb-prisma) by @dengfuping in #21
• chore(bindings): update libseekdb S3 URL prefix to c1a508a4 (seekdb 1.2.0) by @dengfuping in #32

v1.1.1

• Release date: 2026-02-09
• Version: V1.1.1

New features

• The @seekdb/default-embed package is no longer installed automatically with seekdb-js. You must install it manually when using the default embedding function. Run npm install @seekdb/default-embed before use. For details, see Default embedding function.

Bug fixes

• Fixed an issue where the dimension was read incorrectly when deserializing embedding functions.
• Fixed compatibility issues when using embedding functions in both ESM and CJS.

v1.1.0

• Release Date: 2026-01-30
• Version: v1.1.0

📌 Highlights

• More flexible Collection naming: Maximum length increased from 64 to 512 characters, supporting more complex business scenarios.
• Enhanced full-text search: Introduces FulltextAnalyzerConfig, enabling custom analyzers and field mappings.
• Major expansion of the embedding ecosystem: Integrates 11 mainstream embedding providers, covering both open-source and commercial models.
• Upgraded metadata persistence: Embedding function information can be persisted into system tables, improving developer usability.
• New table-fork capability: Supports Collection-level snapshot copying, simplifying experiments and rollbacks.

🆕 New Features

1. Extended Collection name length

• Relaxed limit: Collection names now support 64–512 characters to accommodate longer naming conventions.
• Compatibility: Existing collections created with shorter names are unaffected.

2. Full-text search configuration (FulltextAnalyzerConfig)

• New configuration structure: The createCollection / getOrCreateCollection APIs can accept a configuration parameter. In addition to specifying HNSW index settings, you can now customize full-text index settings via FulltextAnalyzerConfig.

3. Expanded embedding function ecosystem

• Integrates 11 embedding functions across the following platforms:
  • Open-source models: sentence-transformers
  • Commercial services: OpenAI, Azure, AWS Bedrock, etc.

5. Table-Fork feature

• Quickly create Collection snapshots.
• Use cases:
  • A/B testing different index parameters
  • Data rollback and disaster recovery
• Performance impact: Fork is implemented with storage-level copy-on-write (COW) and completes in < 1 second (independent of data size).