v1.3.0 — Smarter Search & Dataset Indexing

Release Date: 2026-07-12
Changes: v1.3.0 → main

Summary

This release focuses on making search more accurate and dataset management easier. It introduces a new optional dataset index that groups documents into topic clusters and provides short overviews, plus improved ingestion and filtering tools so you get better answers faster from your memory data.

Highlights

• New optional "Topic Index" for datasets — groups documents into topic clusters and a short overview to improve search context and answer quality.
• Batch ingestion improvements — easier and faster bulk uploads with progress and resumable retries.
• Search relevance and ranking upgrades — more accurate, less noisy results for common queries.

Breaking Changes

• No breaking changes in v1.3.0. Existing APIs and client integrations should continue to work as before. The new Topic Index and embedding configuration options are optional and opt-in.

New Features

• Topic Index (optional): A new index that organizes a dataset (the collection of documents you've added) into topic clusters and generates a short overview for each cluster. What it does: adds high-level context to search so results consider both specific document matches and nearby topic summaries. Why it matters: search returns more relevant, coherent answers for broad or vague queries without requiring manual tagging.
• Bulk ingestion API and UI improvements: New endpoints and interface controls to upload many files at once, track progress, and automatically retry failed chunks. What it does: makes importing large collections faster and more reliable. Why it matters: reduces the time and manual work needed to populate Cognee with large datasets.
• Metadata filtering in search: You can now include structured metadata (tags, author, date ranges, custom fields) when performing searches. What it does: narrows search to documents that match metadata constraints. Why it matters: improves precision when your memory contains mixed content (e.g., personal notes vs. product documentation).
• Embedding configuration controls: Exposed settings for embedding model selection and chunk size during ingestion. What it does: lets you choose the embedding model and how documents are split into searchable pieces. Why it matters: gives you a direct way to trade off accuracy, cost, and speed for your workflows.

Improvements

• Search ranking improvements: Adjusted scoring so shorter, more relevant passages appear above longer, loosely related matches. This reduces noisy search results and helps quickly surface the best answers.
• Query-time context expansion: When the Topic Index is enabled, search will incorporate short topic overviews to provide broader context. This leads to better responses for queries that need general orientation rather than a single exact match.
• More robust ingestion pipeline: Improved retry logic and chunk handling to prevent dropped or duplicated records during large imports.
• Cleaner error messaging and diagnostics: Search and ingestion errors now return clearer, actionable messages and request IDs to help troubleshooting.
• Updated documentation and quick-start guides: New walkthroughs for creating Topic Indexes, using bulk ingestion, and applying metadata filters.

Performance

• Faster ingestion of large files: Parallelized chunk uploads and improved batching reduce time to ingest large datasets by up to 40% in common scenarios.
• Lower memory during indexing: Internal streaming changes reduce peak memory usage when building indexes, making local deployments more stable on smaller machines.

Security

• Stronger token handling for hosted API keys: Improved token validation and rotation guidance so API keys are less likely to be misused. This improves overall security posture for hosted instances.

Bug Fixes

• Fixed duplicate search results when the same document chunk matched multiple index layers.
• Resolved occasional crash when searching an empty dataset (returned a clear empty-result response instead).
• Fixed pagination bug in dataset listing that could skip or repeat items at page boundaries.
• UI: prevented a freeze when previewing very large documents in the file viewer.
• API: corrected inconsistent timestamps returned by the dataset metadata endpoints.

Technical Changes

• Refactored background indexing workers for improved stability and observability, making long-running jobs easier to monitor and debug.
• Upgraded several internal dependencies to reduce maintenance and improve compatibility with newer runtimes.
• Improved logging and request tracing to aid support and troubleshooting.

Compatibility

— The Cognee Team · 2026-07-12