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