v1.0.0

Release Notes - v1.0.0

Release Date: 2026-04-11
Changes: v1.0.0.dev0 → main

---

Summary

Cognee 1.0.0 is the project's first stable release. It primarily formalizes the release (version bump), enables caching by default for better out-of-the-box performance, and includes test and packaging cleanups to improve reliability and reproducible installs.

Highlights

• Official 1.0.0 release — stable baseline for production use.
• Caching is now enabled by default for improved performance.
• Updated tests to reflect the new caching default (stability improvement).
• Packaging lockfile cleanup to reduce noisy metadata in distributions.

Breaking Changes

• Default caching behavior changed: CacheConfig now has caching=True by default. If your code relied on caching being disabled by default, you must explicitly disable it when creating the configuration (for example, pass caching=False when instantiating CacheConfig or set your configuration/environment accordingly).

New Features

• Formal release: project version bumped to 1.0.0 in pyproject.toml — marks a stable baseline for users and downstream packages.

Improvements

• Caching enabled by default in CacheConfig: users get improved performance and faster responses out of the box without additional configuration.
• Tests adjusted to reflect the new default caching behavior, reducing false negatives in CI and making test expectations match runtime behavior.

Bug Fixes

• Fixed a unit test that asserted the old caching default. The test now expects caching to be True by default, preventing a failing test when running the test suite against the released package.

Technical Changes

• uv.lock was cleaned up to remove revision/upload-time metadata from package entries — internal packaging/lockfile tidy-up that shouldn't affect runtime behavior but may change lockfile diffs and reproducibility tooling.
• Minor test file update: cognee/tests/unit/infrastructure/databases/cache/test_cache_config.py updated to align expectations with the new default.

---

Generated by Cognee Release Notes Generator on 2026-04-11

v1.0.0.dev0

Release Notes - v1.0.0.dev0

Release Date: 2026-04-11
Changes: v1.0.0.dev0 → dev

---

Summary

This is the first developer preview for Cognee (v1.0.0.dev0). It provides the core memory platform foundation — a searchable, structured memory store with a web UI, API access, and SDKs — intended for early testing and integration. Expect experimental APIs and rapid iteration; feedback is welcome.

Highlights

• Initial developer preview: core memory store + web UI
• Public API and SDKs (Python, JavaScript) for programmatic access
• Natural-language search and embedding-backed retrieval
• Improved sync and reliability compared to prototypes
• Export / backup options and basic access controls

Breaking Changes

• This is a developer preview release: APIs and SDKs are considered experimental and may change in future releases. If you build on top of v1.0.0.dev0, be prepared to update integrations when the public stable v1.0.x is released.
• Data model and export format may be updated in future releases. We recommend keeping backups (use the Export feature) and avoiding irreversible production migrations against this dev version.
• Authentication defaults: In this dev preview some auth flows (e.g., API key vs. OAuth) may be enabled or disabled by default depending on install. Check your deployment's auth configuration after upgrading.

New Features

• Core memory store: Save, update, search, and delete memory items with structured metadata and free-text content.
• Web application: A lightweight, responsive UI to browse memories, run searches, and manage collections.
• Public REST API: Programmatic access to create and query memories for integrations and automation.
• SDKs: Official Python and JavaScript SDKs to simplify using the API from applications and scripts.
• Natural-language search: Use plain-language queries to find relevant memories; powered by embedding-based retrieval.
• Export & backup: Download your memory data in standard JSON format for local backup or analysis.
• CLI tool: Basic command-line interface for common tasks like importing/exporting and quick queries (developer-focused).

Improvements

• Search relevance: Better ranking of results by combining keyword matches with semantic embeddings for more helpful results.
• Sync reliability: Reduced risk of duplicate items and improved conflict handling when multiple clients update the same memory.
• Performance: Faster response times for common queries and list operations in the web UI and API.
• Usability: Cleaner UI flows for creating memories, editing metadata, and managing collections.
• Security & auth: Added API key and OAuth support to secure programmatic access (suitable for development and testing environments).

Bug Fixes

• Fixed memory duplication that could occur during rapid repeated saves from external clients.
• Resolved timestamp inconsistencies that sometimes displayed incorrect created/updated times in the web UI.
• Fixed edge-case search regressions where short queries returned no results despite matching content.
• Addressed several UI layout issues affecting smaller screens and narrow viewports.

Technical Changes

• Introduced a plugin-friendly architecture to simplify future integrations (embeddings providers, storage backends).
• Database migrations and initial schema for memory items, collections, and permissions.
• Refactored search pipeline to separate embedding generation from indexing, enabling easier swapping of embedding providers.
• Upgraded several internal dependencies for stability and security; developers should review dependency changes for compatibility in custom deployments.

---

Generated by Cognee Release Notes Generator on 2026-04-11

v0.5.5.dev1

Release Notes - v0.5.5.dev1

Release Date: 2026-04-10
Changes: v0.5.8rc1 → v0.5.5.dev-fix-baseten

---

Summary

This is a small maintenance pre-release addressing a Baseten-related runtime issue and improving robustness when LiteLLM model metadata is incomplete. The release adds defensive checks and clearer logging for model token limits, bumps the package version to a development pre-release, and includes lockfile / packaging metadata updates to reduce install-time issues on some platforms.

Highlights

• Fix for a Baseten-related crash caused by missing model metadata (max_tokens).
• Improved robustness in LLM utilities: the code now checks for missing model fields and logs helpful diagnostics instead of failing.
• Version bumped to 0.5.5.dev1 (development pre-release).
• Packaging/lockfile updates to reduce installation problems (sdist entries added and platform markers to avoid incompatible pyobjc installs on certain Python/OS combos).

Breaking Changes

• No breaking changes to the public API. Upgrading should be safe for existing users.
• Note: This is a development pre-release (0.5.5.dev1). Packaging/lockfile changes may alter which optional native/macOS-only packages are installed on Windows Python 3.12 — this prevents install-time failures but may change which platform-specific packages appear in your environment.

Improvements

• get_model_max_completion_tokens (cognee/infrastructure/llm/utils.py) now safely handles models that don't expose a "max_tokens" field in LiteLLM's model_cost; it logs a clear message instead of raising an error.
• Debug logging improved to make it easier to diagnose missing model metadata when connecting to hosted model providers like Baseten.
• pyproject.toml updated to mark this build as 0.5.5.dev1 (development pre-release).
• uv.lock updated: several sdist entries were added and dependency markers were introduced so some pyobjc-related packages are not installed on Windows with Python 3.12, reducing platform-specific install failures.

Bug Fixes

• Resolved the Baseten issue where Cognee could encounter errors when LiteLLM's model_cost lacks a "max_tokens" entry—Cognee now avoids the crash and logs the missing information instead.

Technical Changes

• Code: defensive check added in cognee/infrastructure/llm/utils.py to verify presence of the "max_tokens" key in LiteLLM's model_cost before using it.
• Metadata: pyproject.toml version changed to 0.5.5.dev1.
• Lockfile: uv.lock updated with sdist entries and platform markers for a few packages to improve cross-platform install behavior.

---

Generated by Cognee Release Notes Generator on 2026-04-10

v0.5.7.dev0

Release Notes - v0.5.7.dev0

Release Date: 2026-04-09
Changes: v0.5.8rc1 → dev

Pull Requests: #2126

---

Summary

This release continues the v2 memory-first API work and improves local and cloud developer workflows. It introduces major session-memory features (session-only remembers/search, automatic session search on prompts), a new public SaaS frontend subset, cognee.serve() to connect the SDK to Cognee Cloud (and run a local instance), and various reliability, performance, and pipeline improvements.

Highlights

• New session-first memory behaviors: session-only remembers, session cache full-text storage, and automatic session search on every prompt.
• cognee.serve() to connect your SDK to Cognee Cloud or run local instances; cognee.visualize now exposed in the v2 API.
• Public subset of a new SaaS frontend included (updated UI + assets) for easier local cloud-like testing.
• Performance: lazy imports -> ~7.6x faster startup; pipeline improvements including parallel execution and concurrency limiting.
• Breaking change: cognee.status() and related freshness/status hooks were removed.

Breaking Changes

• Removed cognee.status(): code or tooling that relied on cognee.status() or its freshness checks must be updated — references to this API and related checks have been removed.
• Frontend changes: the old cognee-frontend demo/example assets were replaced with a new SaaS frontend subset. If you have local customizations that depended on the previous frontend files (assets, demo data, or routes), you'll need to update them to match the new layout and asset set.
• Default local user email changed to local-ui@cognee.local. If you relied on the previous default email in scripts or tests, update them accordingly.

New Features

• Session memory enhancements: remember() now supports separate session-only and permanent memory paths, session IDs are unified, and session-only recall is supported.
• Automatic session search on every prompt via the UserPromptSubmit hook so relevant session memory is surfaced without manual steps.
• Session cache improvements: store full text, support multiple entries per session, and carry session_ids in RememberResult for easier traceability.
• cognee.serve(): connect the SDK to Cognee Cloud or run against a local instance (local instance support added).
• cognee.visualize exposed in the v2 API as cognee.visualize for programmatic dataset visualization.
• New self-improvement pipeline and graph-to-session sync; RememberResult can be returned as a promise and is displayed in the CLI and Claude plugin.
• New simplified pipeline patterns and declarative TaskSpec/BoundTask/run_pipeline features for deferred-call pipelines.

Improvements

• Frontend: replaced old cognee-frontend with a new SaaS frontend public subset — updated pages, components, assets, and onboarding flows for a cloud-first developer experience.
• Filesystem cache TTL support added and max_context_chars wired up so cached entries can expire and context size is enforced.
• Example apps revised: clearer permanent vs session memory patterns, per-source status and freshness checks, and run_migrations() to make example runs repeatable.
• Pipeline engine: added concurrency limiting, incremental loading, parallel execution, and per-item context injection in parallel mode for better throughput and scalability.
• Default local user email changed to local-ui@cognee.local to simplify local sign-in flows.
• Startup performance improved by ~7.6x via lazy imports in cognee/init.py.
• Claude plugin and CLI improved with session/feedback args and updated scripts for session sync and search.

Bug Fixes

• Multiple CI/test stability fixes: increased timing thresholds in flaky tests, improved imports/patching in tests, and MagicMock serialization fixes.
• Lint and formatting fixes (ruff) and removal of stray trailing whitespace/newlines in multiple files.
• Fixed usage of create_db_and_tables for fresh databases in examples and enabled filesystem caching in examples to support session behavior.
• Fixed word-boundary session search, unified session_ids across features, and resolved various import-order issues.
• Removed leftover references to a removed freshness check to avoid runtime errors.

Technical Changes

• Removed dead code including the unused Pipe[T] subsystem and the old cognee.status implementation to simplify the codebase.
• Library upgrades: litellm bumped to 1.83.1 and graphiti-core to >=0.28.0.
• Refactoring and internal API changes across the v2 memory-oriented API (remember, recall, improve), pipeline infrastructure, and task execution patterns.
• Many internal test and CI harness changes (imports, mocks, and test context migration to PipelineContext) to improve reliability.

---

Generated by Cognee Release Notes Generator on 2026-04-09

v0.5.4.dev3

Release Notes - v0.5.4.dev3

Release Date: 2026-04-09
Changes: v0.5.8rc1 → v0.5.4.dev3-baseten

---

Summary

This release (0.5.4.dev3) fixes a runtime issue with fetching model token limits from LiteLLM/Baseten model metadata and improves related debug logging. It also updates the package dev version. No user-facing API changes — upgrade recommended if you encountered crashes related to model token metadata.

Highlights

• Fix for Baseten/LiteLLM integration that prevents crashes when model metadata is missing token limits.
• Improved debug logging in the LLM utilities to make diagnosis easier.
• Development version bumped to 0.5.4.dev3.

Improvements

• get_model_max_completion_tokens now safely handles models that do not include a "max_tokens" entry in LiteLLM's model_cost, avoiding KeyError crashes.
• Added clearer debug logs when the model entry is present but missing max_tokens, and when a model isn't found at all — makes troubleshooting easier when using external model providers like Baseten.

Bug Fixes

• Resolved issues with Baseten usage by guarding access to the "max_tokens" field in LiteLLM's model_cost. This prevents errors when model metadata doesn't include token limit information.

Technical Changes

• Small code change in cognee/infrastructure/llm/utils.py to check for the presence of the max_tokens key before reading it.
• Version bump in pyproject.toml and uv.lock from 0.5.4.dev1 to 0.5.4.dev3.
• No public API changes. Internal robustness and logging improvements only.

---

Generated by Cognee Release Notes Generator on 2026-04-09

v0.5.8rc1

Release Notes - v0.5.8rc1

Release Date: 2026-04-08
Changes: v0.5.8rc1 → feat/sales-benchmark-demo

---

Summary

Release candidate 0.5.8rc1 introduces the new Sales Benchmark Demo (preview) and a set of relevancy, ingestion, and stability improvements aimed at making memory-driven workflows easier to test and evaluate. This RC focuses on user-facing demo functionality, faster and more accurate memory retrieval, and a handful of bug fixes and internal improvements to prepare for general availability.

Highlights

• Sales Benchmark Demo (preview): an end-to-end example to evaluate Cognee's memory and retrieval on sales-related data.
• Improved memory ingestion and retrieval accuracy for more relevant results and faster query responses.
• New demo dashboard and sample prompts to accelerate onboarding and evaluation.
• Stability and performance fixes for ingestion, sync, and API responsiveness.

Breaking Changes

• No breaking changes in this release candidate. Existing integrations and APIs remain compatible. If you are using beta or internal endpoints, continue as before — no action required.

New Features

• Sales Benchmark Demo (preview) — A guided demo that shows how to ingest sales data, run benchmarking queries, and evaluate retrieval relevance. Ideal for teams who want a hands-on example of Cognee in a sales context.
• Demo dashboard and example prompts — A simple UI and ready-made prompts to help you explore the demo quickly and reproduce sample workflows without building queries from scratch.
• Benchmark ingestion pipeline — A pre-configured ingestion path that demonstrates best practices for structuring sales records and metadata to get the best retrieval results.

Improvements

• Faster and more accurate retrieval — Relevance ranking and indexing have been tuned to return more useful memory results for typical sales queries, reducing noise in answers.
• Improved ingestion robustness — Better error handling and retry behavior during bulk imports to reduce failed records and partial ingestion states.
• Long-context handling — Memory retrieval is more resilient when prompts include longer context windows, reducing truncation-related mismatches.
• Cleaner onboarding experience — Example prompts, improved demo walkthrough, and clearer UI copy to help new users get up and running faster.
• API ergonomics — Subtle improvements to response payloads and pagination to make programmatic consumption smoother.

Bug Fixes

• Fixed an intermittent race condition during concurrent ingestion that could cause duplicated or missing records.
• Resolved a UI layout glitch in the demo dashboard that caused truncated text in certain browser sizes.
• Fixed pagination edge-case in the API that could return an empty page when results spanned multiple pages.
• Addressed a memory sync issue where recently added items might not be immediately available to very recent queries.

Technical Changes

• Refactored parts of the memory indexing core to simplify future feature development and improve maintainability.
• Upgraded several internal dependencies and improved test coverage around ingestion and retrieval logic.
• Internal logging and telemetry enhancements to make it easier to investigate ingestion and retrieval issues during the demo.

---

Generated by Cognee Release Notes Generator on 2026-04-08

v0.5.8

Release Notes - v0.5.8

Release Date: 2026-04-07
Changes: v0.5.8 → main

---

Summary

Release 0.5.8 is a stability and usability focused update that sharpens search relevance, improves memory management workflows, and fixes a variety of issues reported since v0.5.7. This release prioritizes reliability and day-to-day productivity—most users can upgrade safely without any changes to their integrations.

Highlights

• Improved semantic search relevance for faster, more accurate memory recall
• New memory tagging and batch-edit features to manage large collections of memories
• Performance and reliability improvements across the UI and backend
• Multiple bug fixes addressing sync, pagination, and export/import issues

Breaking Changes

• No breaking changes in this release. Existing integrations, API clients, and configurations should continue to work as before. We recommend testing upgrades in a staging environment for mission-critical systems.

New Features

• Memory tagging: add, edit, and filter by tags to organize memories more effectively—useful for grouping related notes and speeding retrieval.
• Batch editing and bulk actions: perform operations (tagging, deleting, exporting) on multiple memories at once to save time when cleaning or organizing data.
• Improved export/import flows: more robust handling of large exports and clearer progress indications during imports to avoid interruptions.

Improvements

• Search relevance tuning: ranking improvements result in more accurate top results for common queries—fewer manual searches to find what you need.
• Faster memory recall: backend optimizations reduce latency for common read and search operations, improving responsiveness in the app.
• UI polish: clearer empty states, better error messages, and smoother interactions in list views and detail pages to reduce user friction.
• Sync robustness: improved conflict resolution and retry logic reduces chances of lost or duplicated memories during intermittent connectivity.

Bug Fixes

• Fixed issue where pagination could skip or repeat items in some list views.
• Resolved occasional duplicate memories appearing after quick successive imports.
• Fixed crash when uploading very large files or very long text fields during import/export.
• Addressed OAuth refresh issues that caused some integrations to stop syncing automatically.
• Corrected several UI glitches in the memory detail and list views (layout and action button states).

Technical Changes

• Internal refactor to the memory indexing pipeline to support future enhancements and cleaner performance tuning.
• Upgraded several dependencies for security and stability (minor version bumps); compatibility-tested with existing deployments.
• Improved unit and integration test coverage around sync, import/export, and search components to reduce regressions.

---

Generated by Cognee Release Notes Generator on 2026-04-07

v0.5.7

Release Notes - v0.5.7

Release Date: 2026-04-03
Changes: v0.5.7 → main

---

Summary

Cognee 0.5.7 focuses on making memories faster, more accurate, and easier to work with. This release improves search relevance and sync speed, adds richer attachments and sharing options, and fixes a number of reliability issues—while introducing one important format change for memory exports.

Highlights

• Significantly improved semantic search relevance and speed
• Support for multi-file and richer attachments inside memories
• Workspace sharing and access controls for smoother collaboration
• Faster initial sync and reduced memory footprint on clients
• Important export format change (breaking) — see Upgrade notes

Breaking Changes

• Memory export format updated to v2: The JSON shape for exported memories has changed (field names and structure). If you use automated imports/exports or third-party tools, update them to handle the new export format. We provide a compatibility tool for converting v1 exports—run the migration or re-export from the app.
• API note: Some internal API fields used by integrations were renamed; most official SDKs have been updated. If you maintain a custom integration against the internal API, please update to the latest SDK or review the API change log.

New Features

• Multi-file attachments: You can now attach multiple files (documents, images, PDFs) to a single memory and view previews inline—making it easier to store and recall context-rich items.
• Workspace sharing & roles: Share memory collections with teammates and assign basic roles (viewer/editor) to control access without leaving the app.
• Memory undo for deletes: A 30-day “trash” holds deleted memories so you can restore them if removed by mistake.

Improvements

• Search relevance and ranking: Updated embedding and re-ranking logic yields more accurate search results for short and long queries; users will notice more useful top results.
• Faster initial sync: Clients now perform a smaller, prioritized sync on first startup, reducing time-to-usable.
• Reduced client memory usage: Background caching and lazy-loading reduce RAM use for large workspaces.
• Improved attachment handling: Previews and faster uploads for common file types; larger file uploads are handled more robustly.
• UI polish: Cleaner memory list, clearer breadcrumbs, and improved empty-state guidance to help new users get started.

Bug Fixes

• Fixed intermittent duplicate memories appearing after sync completion.
• Resolved an issue where long-running searches could time out and return stale results.
• Fixed auth token refresh bugs that caused unexpected logout for some users.
• Corrected broken links in shared workspace invites and resolved inconsistent permissions display.
• Resolved a crash when saving very large attachments in some browsers.

Technical Changes

• Storage refactor: Server-side storage layer refactored to improve scalability and prepare for sharding. This should be transparent to users but reduces future downtime risk.
• Dependencies upgraded: Key backend libraries updated for performance and security improvements.
• Telemetry improvements: Better instrumentation for search and sync to help diagnose future issues faster.

---

Generated by Cognee Release Notes Generator on 2026-04-03

v0.5.6

Release Notes - v0.5.6

Release Date: 2026-03-30
Changes: v0.5.6 → main

---

Summary

Cognee 0.5.6 focuses on making memories faster, more relevant, and easier to manage. This release delivers a small set of user-facing features—bulk import/export and better relevance controls—plus search and stability improvements to reduce friction for both end users and teams.

Highlights

• Faster memory retrieval and search responsiveness for a smoother experience
• Smarter relevance ranking so results are more context-aware
• Bulk import/export for memories to simplify migration and backups
• New memory filtering and tagging improvements in the UI
• Multiple bug fixes and stability improvements across sync, auth, and dashboard

Breaking Changes

• No breaking changes in 0.5.6. Integrations and existing workflows should continue to work as before. If you rely on the API, review the compatibility notes in the API docs as we’ve made small, non-breaking improvements to response formats and error messages.

New Features

• Bulk import and export: You can now export a set of memories or import large batches to/from JSON/CSV formats. This simplifies backups, migrations, and onboarding large datasets.
• Manual tagging & labels: Add custom tags and labels to memories to improve organization and make filtering more effective.
• Relevance controls: New settings let you tune relevance signals used in memory ranking so results match your context and priorities better.
• Memory details view: The UI includes an expanded details panel for individual memories with metadata, tags, and recent activity.

Improvements

• Search & retrieval performance: Reduced latency for common queries and improved pagination to return results faster.
• Deduplication & merging: Improved logic to detect and merge duplicate memories, reducing clutter and confusion.
• UI filters: Faster, more reliable filters (by tag, date, and source) and clearer empty-state messaging.
• API behavior & errors: API responses have clearer error messages and more consistent pagination defaults to make integrations more reliable.
• Sync reliability: Reduced edge-case failures during syncing with external sources (fewer dropped or duplicated items).

Bug Fixes

• Fixed an issue where importing certain memory batches could create duplicate entries.
• Resolved a rare authentication token refresh failure that caused unexpected logouts.
• Fixed a crash that could occur when deleting archived memories from the dashboard.
• Corrected inaccurate memory counts in some dashboard widgets and lists.
• Addressed a problem where empty fields could cause memory creation to fail in some workflows.

Technical Changes

• Internal refactor of the memory storage layer to improve future scalability and maintainability.
• Upgraded several backend dependencies for security and performance.
• Improved test coverage and monitoring to detect regressions earlier.
• Telemetry enhancements to better track query performance and sync health (privacy-respecting; no new user-visible data collection).

---

Generated by Cognee Release Notes Generator on 2026-03-30

v0.5.5

Release Notes - v0.5.5

Release Date: 2026-03-14
Changes: v0.5.5.dev0 → main

Pull Requests: #2288

---

Summary

Release 0.5.5 brings a major push around DLT ingestion (new MVP, safety and robustness), an important memify change to use triplet embeddings by default, and multiple CI/docs automation and bug fixes. This release focuses on enabling reliable ingestion from DLT sources, improving developer workflows, and tightening up stability across examples and tests.

Highlights

• DLT ingestion MVP for Postgres with full supporting modules and example usage
• Memify default pipeline now uses triplet embeddings for better semantic matching
• Hardened DLT ingestion: SQL safety, async handling, concurrency controls, and error handling improvements
• New CI/docs automation: install_cognee action, more flexible test workflows, and automatic docs sync on releases
• Numerous bug fixes and Windows Docker entrypoint and Swagger UI fixes

Breaking Changes

• Memify default pipeline now uses triplet embeddings instead of coding rules — this can change embedding vectors, similarity results, and downstream retrieval behavior. Review any custom memify pipelines or expectations when upgrading.
• DLT ingestion default write disposition changed to 'replace'. Existing destinations may be overwritten by ingestion runs unless explicitly configured otherwise. Double-check your DLT ingestion settings before running against production destinations.
• Destination creation for DLT ingestion now uses dlt_db_name instead of db_name in relational configs — users who relied on the old db_name behavior may need to update configurations.
• Releases created from the 'dev' branch are now marked as pre-releases by automation; if you rely on non-prerelease tags from dev, adjust your workflow.

New Features

• DLT ingestion support: added ingestion pipeline, source resolution, destination creation, FK extraction, and example demo (Postgres destination)
• Memify: replaced coding rules with triplet embedding in the default memify pipeline to improve embedding quality and retrieval
• Docs automation: sync Mintlify docs on published releases and a new tool to sync OpenAPI and changelog (tools/sync_release_docs.py)
• New GitHub composite action to install Cognee in CI (supports local install or specific PyPI version)

Improvements

• DLT ingestion hardened for SQL safety, async execution, concurrency, and orphan row logic
• DLT ingestion usability: resolve relative SQLite paths, recognize DltSource types, use dlt_db_name for destination creation, and default write disposition set to 'replace'
• Validated file paths and surfaced errors in the MCP cognify tool to avoid silent failures
• Added DLT_MAX_ROWS_PER_TABLE to .env.template (commented) to control how many rows are read per table during DLT ingestion
• CI improvements: CLI and MCP test workflows support installing a specific Cognee version and use the new install_cognee action; dev releases are marked as pre-releases in automation
• Repository hygiene: enforced LF for shell scripts (.gitattributes) to fix Docker entrypoint issues on Windows, and various code formatting and lockfile updates

Bug Fixes

• Fixed Swagger UI issue and several test assertion failures
• Guarded against None tracer in new_span() to avoid crashes in observability code
• Fixed Docker entrypoint failing on Windows by ensuring LF line endings and related Dockerfile adjustments
• Fixed many DLT-related ingestion bugs: DataItem ingestion, write disposition handling, import issues, and recognition of DltSource in resolution
• Replaced broad exception handlers with narrower ones in task code, improved NotImplementedError messages, and removed mutable default arguments to avoid subtle bugs
• Fixed community greeting workflow to skip bot/Claude Code PRs and corrected parameter names for actions/first-interaction@v3

Technical Changes

• Consolidated DLT logic into a single resolve_dlt_sources adapter and added many new DLT modules (ingest_dlt_source, create_dlt_source, dlt_utils, extract_dlt_fk_edges, etc.) and example data under examples/demos/dlt_ingestion_example.py
• Added tools/sync_release_docs.py to automate OpenAPI and changelog sync to the docs repository on release
• Added .github/actions/install_cognee/action.yml (composite action) and updated multiple GitHub workflows to support passing cognee_version and installing local or specific PyPI versions
• Updated dependencies and lockfiles to resolve vulnerabilities and modernize the test/deploy stacks
• Misc: refactors, type/typing fixes, circular import fixes, and formatting changes across the codebase

---

Generated by Cognee Release Notes Generator on 2026-03-14