v0.2.28

🚀 New Core Capabilities

• Modular Skills Support: We've introduced a framework for "Skills," allowing for a more extensible and granular way to define agent capabilities.
• Advanced Knowledge Engine: The integration of a dedicated Knowledge Engine significantly enhances the agent's ability to process and retrieve domain-specific information.
• Contextual Summarization: Improved summarization capabilities across both SDK and CRM contexts allow the agent to maintain better coherence over long-running interactions.

🛠️ Architectural & Performance Gains

• Faster Embeddings: We’ve replaced the heavy torch/sentence-transformers stack with fastembed, significantly reducing our footprint and improving speed.
• Watsonx & Model Flexibility: Improved JSON parsing for Watsonx and expanded support for diverse models (including gpt-oss-20b) ensure smoother inference across different LLM providers.
• Streamlined Project Structure: We've migrated tools and documentation into a unified src package to simplify the development experience.

🔒 Enterprise Reliability & Security

• Hardened Security: This release addresses over 37 security vulnerabilities (CVEs) across our frontend and backend dependencies, ensuring a more secure environment.
• Vault KV v2 Support: Enhanced HashiCorp Vault integration provides more robust path resolution and secret management for production deployments.
• Production Readiness: New feature flag overrides and refined container build processes ensure more predictable agent behavior on startup.

What's Changed

• feat: replace torch/sentence-transformers with fastembed and make docs MCP server generic by @sami-marreed in #114
• fix: apply feature flag overrides to production agent on startup by @sami-marreed in #116
• chore: bump fastmcp to 3.2.0 for GHSA-vv7q-7jx5-f767 by @sami-marreed in #117
• Feat/knowledge engine by @dolev31 in #125
• fix: reconcile litellm 1.83.4 with optional openlit and UBI Docker by @sami-marreed in #130
• fix: migrate tools and docs to package src by @sami-marreed in #131
• Kaizen lite mode by @gjt-prog in #74
• fix(deps): remediate critical/high CVEs in Python dependencies by @offerakrabi in #134
• Feat/context summarization by @offerakrabi in #61
• fix(deps): upgrade frontend dependencies to fix 37+ security vulnerabilities by @offerakrabi in #139
• fix(security): upgrade dependencies to resolve high-severity CVEs by @offerakrabi in #140
• fix: handle nodes for openlit integration by @sami-marreed in #142
• fix: fix playwright version by @sami-marreed in #143
• chore: Upgrade langchain text splitters package by @erichare in #145
• fix: tag am64 image as a seperate by @sami-marreed in #147
• Update readme by @gaodan-fang in #146
• fix: address prompt issues by @sami-marreed in #148
• feat: improve Vault KV v2 support and path resolution by @sami-marreed in #167
• fix: address test failures in CRM and SDK context summarization by @sami-marreed in #169
• Remove memory support for cuga classic by @gaodan-fang in #153
• fix: refine container build process and entrypoint by @sami-marreed in #171
• fix(task-analyzer): add app name normalization and typo correction by @offerakrabi in #170
• fix(watsonx): use prompt-based JSON parsing instead of response_format by @haroldship in #89
• fix: handle models like gpt-oss-20b by @sami-marreed in #178
• fix: address knowledge file startup by @sami-marreed in #182
• feat: support skills in cuga by @sami-marreed in #121

New Contributors

• @dolev31 made their first contribution in #125
• @gjt-prog made their first contribution in #74
• @erichare made their first contribution in #145
• @gaodan-fang made their first contribution in #146

Full Changelog: v0.2.11...v0.2.28