Skip to content
Agents tracked: 442 Downloads (7d): 202M down 19.3% GitHub stars: 10.4M Releases (7d): 693 Agent status: 1 with issuesSince last limit reset: Claude 28d · Codex 1d Updated Oct 21, 2026
AgentGid
✦ Ask

Cognee changelog: what's new each month

Every stable Cognee release summarised by month: the highlights, new features, improvements, fixes and anything you need to act on. 3 months covered; the current month updates daily.

October 2026 · so far

3 releases: 1.6.3 → 1.6.5

October brought improvements to search accuracy and stability across Cognee, with new memory ingestion capabilities and better handling of vector data. The month focused on reliability fixes, restricted agent nesting, and optimized database operations.

Action needed
  • Removed temporalcognify configuration option — remove references from your configuration.
  • Agents can no longer create nested agents — update any automation or scripts relying on this.
  • LanceDB automatic background pruning removed — you must run compaction manually if needed for storage reclamation.
  • datamodel-code-generator requirement bumped to >= 0.56.0 if you use custom graphmodel generation.

Highlights

  • Natural-language searches now return matching database rows directly and skip unnecessary analysis steps.
  • Temporal searches now include boundary dates and rank date matches more accurately.
  • New Memify API endpoints let you convert content into memory entries programmatically.
  • Code docstrings are now stored as graph documents and linked to code symbols for better recall.
  • Search memory usage reduced by excluding raw vector columns from LanceDB and pgvector results.

New

  • Memify API endpoints for converting content into memory entries via the API.
  • Base64 embeddings support for better compatibility with embedding providers.
  • Code-graph docstrings stored as documents and linked to code symbols.
  • Daily CVE Slack alerts for security monitoring.
  • Google Calendar connector with improved OAuth and token handling.

Improved

  • Temporal search windows now include their boundary dates.
  • Trace processing is faster with fewer, larger database operations.
  • Ingestion validation strengthened to reject malformed node sets.
  • Notion block rendering fixed with no depth caps.
  • Turso errors now surface as OperationalError for consistent error handling.
  • Docker images now include git for better compatibility.
  • Parent and agent dataset access behavior corrected.

Fixed

  • Natural-language searches no longer include 'turn analysis' that could add unrelated context.
  • Vector column no longer returned in LanceDB and pgvector searches, preventing memory failures.
  • DLT dependency pinned below 1.31 to avoid import errors.
  • LanceDB background pruning removed to stop interference with searches.
  • Intermittent subprocess and session failures reduced through test and harness fixes.
  • Agent creation restrictions corrected so agents cannot create nested agents.

September 2026

4 releases: 1.5.4 → 1.6.2

September brought major usability improvements to Cognee, including a keyless-first experience for local models, new Google and Slack connectors, and significant stability enhancements across data ingestion, pipelines, and embeddings. The month focused on making the system work smoothly without requiring cloud LLM keys and simplifying data import from common enterprise sources.

Action needed
  • Removed 'repositories' parameter from affected API/CLI calls; update integrations to the new calling pattern.
  • COGNEEALLOWEDLOCALFILEROOTS configuration is now disabled by default; explicitly re-enable if you previously used local file access.
  • GLiNER no longer bundled in default Docker image; add it to your image or installation if needed.
  • Skills moved from earlier locations to .agents/skills; update custom scripts, automation, build hooks, or CI that reference old paths.

Highlights

  • Keyless-first experience: many workflows now run cleanly without an LLM API key, with clear announcements when local model downloads are needed.
  • New Google Drive OAuth connector with Gmail and shared-drive sync for secure, streamlined imports from Google Workspace.
  • Slack integration lets you import and continuously sync conversation history into Cognee datasets.
  • Pipeline robustness improved: runs now recover reliably after crashes while preserving completed documents.
  • Embedding fixes prevent silent truncation and ensure consistent embeddings by respecting each model's token limits.

New

  • Google Drive OAuth connector for authorized file sync from Google accounts and shared drives.
  • Gmail ingestion and shared Google Drive sync support with progress indicators.
  • Slack integration to import and continuously sync conversation history.
  • API add endpoint now accepts rawdata strings and git repo URLs for easier programmatic dataset ingestion.
  • cognee-mcp client/server and CLI improvements for Enola integration.
  • Streaming /visualize/json responses in chunks to handle large graph views without memory exhaustion.
  • Document external metadata now carried in chunks and surfaced in hybrid retrieval results.
  • GLiNER auto-installer that runs off the event loop and installs CPU Torch on first use.

Improved

  • Docker image now includes the dlt extra by default so common data-ingestion tools work out of the box.
  • Frontend reorganized with numerous UI fixes improving dataset, dashboard, and integration workflows.
  • Pipeline runs are stamped at start and recover reliably after crashes, preserving completed documents.
  • Model wiring ensures image and audio transcription models are passed through end-to-end with embedding models recorded per dataset.
  • Dataset names fingerprinted before leaving the host and error telemetry reports full routes for better diagnostics.
  • Chunks sized by each model's token limit to ensure content is embedded intact and consistently.
  • Better error messages and faster failure detection on unrecoverable embedding errors.
  • CLI and documentation refreshed including database migration commands and Docker quickstart auth fixes.

Fixed

  • Multiple fixes to migrations, logging, and provenance to improve stability during upgrades and concurrent operations.
  • Embedding fixes prevent silent truncation of content.
  • Skills recovery implemented when file linking fails.
  • Security hardening: API tokens are now redacted in startup logs.
  • Local file-root access disabled by default for improved security.
  • Windows linking support added for skills directory.

August 2026

5 releases: 1.4.2 → 1.5.3

August brought significant improvements to search relevance and data ingestion reliability across multiple releases. The month focused on making Cognee faster and more stable, with refinements to how results are ranked and how documents are processed.

Highlights

  • Search now returns more relevant results for short and ambiguous queries by using broader contextual signals and improved ranking.
  • Document ingestion is faster and more reliable, with the ability to resume interrupted uploads and better handling of large files.
  • Admin controls let you manage dataset indexing, including starting, pausing, and re-running indexing operations.
  • LLM outputs can now be made reproducible with temperature and seed support where providers allow it.
  • Windows support improved with bundled OpenSSL to resolve TLS connectivity issues.

New

  • Optional smart indexing mode that groups datasets into topic clusters and generates summaries for each cluster.
  • Admin interface controls to start, pause, and re-run dataset indexing operations.
  • SDK convenience helper functions for common ingestion and query workflows.
  • Resumable ingestion: interrupted uploads and indexing jobs can now continue from where they stopped.
  • Community plugins documentation and a tutorial for migrating from other systems like Letta, MemGPT, and Zep.
  • Improved file validation during upload to catch malformed or unsupported files earlier.
  • Large-scale migration testing in CI to verify data migration reliability.

Improved

  • Search ranking algorithm tuned to handle short queries better and boost clearer, more relevant matches.
  • Ingestion progress indicators show clearer feedback and more actionable error messages.
  • Document upload process is more resilient to interruptions and transient errors.
  • Error messages throughout ingestion and search workflows are clearer and more helpful for debugging.
  • Graph adapter performance improved with set-based operations, chunked queries, and batched processing.
  • Logging and observability enhanced for better visibility during ingestion and indexing.
  • Search now uses broader contextual signals to return more relevant results.
  • LLM temperature control and seed support added for reproducible outputs.

Fixed

  • Multiple crash and duplication issues that affected long-running projects.
  • Sync failures and internal stability problems reduced through hardening.
  • Common upload interruptions no longer leave partial or duplicate entries in datasets.
  • TLS issues on Windows resolved with bundled OpenSSL.
  • Graph migration reliability and correctness improved with fixes to fork-splitting and rekey operations.

Summaries are written automatically from the official release notes (full changelog ↗); check the original notes before relying on a detail. Cognee: pricing, features and alternatives · All changelogs