New Features:
- Generalist Agent Revamp: Major overhaul of the generalist agent experience, including step-by-step reasoning API, task grouping in the WebUI, persistent agent memories, and improved orchestration for tool calls and sub-tasks.
- MCP Client Integration: The MCP client has been relocated into the Agents service, providing unified endpoints for listing and invoking MCP tools with robust HTTP/SSE transport detection, retries, and structured result parsing.
- Session and Tool Duration Tracking: Activity logs now display total session duration and per-tool execution times, improving observability and debugging.
- Agent Description Field: Agents now support editable descriptions, with auto-generation for missing values, enhancing discoverability and clarity in the UI.
- OAuth2 CLI Login Flow: Streamlined OAuth2 login for CLI users, with automatic API key selection and secure callback redirects.
Improvements:
- WebUI and Chainlit Enhancements:
- Task grouping and parented tool/sub-task steps in the Chainlit UI.
- Improved streaming, error handling, and security headers for hosted assistants.
- Refined CSS to reduce visual clutter and improve accessibility.
- Loader states and progress indicators for iframe chat initialization and pending tasks.
- Agent Activity Logs:
- Grouped user turns, tool calls, and assistant answers for improved readability.
- Display of input/output LLM token metrics and per-turn durations.
- Robust parsing of tool call payloads and headers for clearer operation tooltips.
- Environment and Deployment Management:
- Unified environment handling with clear default logic and improved sorting.
- Enhanced SQL functions for filtering non-default environments with deployments.
- Model Catalog Expansion:
- Added GPT-5.1 and additional GPT-5 family variants to the model catalog, with updated context and output token limits.
- Authentication and Security:
- Hardened OIDC and WebUI authentication flows with safe redirect handling and improved session recovery.
- Support for passing redirect URLs and metadata through OIDC login and callback flows.
Bug Fixes:
- Async Platform Stabilization:
- Resolved failures in agentic task runs by aligning the scheduler and job executor with async execution models.
- Fixed event-loop conflicts and improved error handling in scheduled jobs.
- Agent Execution and Activity Export:
- Fixed thread export in agent activity by ensuring unique message serialization and returning serializable session data.
- Addressed missing execution records during task updates by enabling upsert-on-missing logic.
- Billing and Quota Calculation:
- Corrected billing quota usage calculations by aligning with entitlement-based usage limits and sunsetting the MCP service in workflows.
- UI Consistency and Minor Fixes:
- Addressed various UI bugs, including modal race conditions, loader states, chat background color, and metrics tab rendering.
- Improved error propagation and PII error handling in activity threads.
- Fixed metrics aggregation and token metrics display in activity logs.
- Ensured robust handling of tool-call reporting, agent error flows, and tool result extraction in the UI.
