November 2025

Development recap for November 2025, highlighting new features, integrations, improvements, and fixes across GL AIP.

1. Security Improvements for Custom Tools

   Enhanced security controls and validation for custom tools to reduce risk, prevent misuse, and ensure safer execution within agent workflows.

2. BUA Improvements

   Improved reliability, performance, and overall usability of BUA to deliver more consistent behavior and smoother user experiences.

3. Migrated from Direct mem0ai to gllm_memory Adapter

   Replaced direct mem0ai usage with the gllm_memory adapter to improve abstraction, maintainability, and long-term extensibility of memory management.

4. Monitoring Agent Runs

   Added monitoring capabilities for agent executions, providing better visibility into run status, execution flow, and potential failure points.

5. Descriptive Tool and Delegate Messages

   Tool and delegate messages are now more descriptive, making agent behavior easier to understand, debug, and audit.

6. Added BOSA Probing Functionality for BOSA Integration

   Introduced probing capabilities to validate connectivity, configuration, and readiness when integrating with BOSA.

7. Added GPT-5.1 Variants

   Expanded model support with GPT-5.1 variants, with configurable reasoning effort to balance response quality, latency, and computational cost.

8. MCP Server Tool Whitelisting

   Implemented tool whitelisting for MCP servers, allowing tighter control over which tools are permitted to run.

9. Added BOSA_API_KEY Configuration Support to tool_config

   Enabled configuration of BOSA_API_KEY directly within tool_config, simplifying setup and improving consistency across environments.

10. Migrated Steel Session Recording Architecture to Headful Mode

    Updated Steel session recording to use headful mode, improving recording compatibility and observability during interactive sessions.