- Update all LLM model files to use getSecretSync with AI_KEYS constants
- Migrate OpenAI GPT models (gpt-5, gpt-5-mini, gpt-5-nano) to use secrets system
- Migrate Anthropic models (sonnet-4, haiku-3-5) to use secrets for both Anthropic API and Vertex AI credentials
- Update provider files (openai.ts, anthropic.ts, vertex.ts) to use getSecretSync
- Migrate evaluation workflow to use BRAINTRUST_KEY from secrets
- Implement proper error handling for missing credentials with try-catch blocks
- Test files will be migrated in a separate update
All non-test source files now use centralized secrets management instead of direct process.env access.
- Added checks to ensure toolCallId exists and is a valid string before processing tool results.
- Preserves the original part if toolCallId is missing or invalid, improving robustness in message conversion.
- Removed the creation and handling of raw LLM message entries in both start and finish functions.
- Updated message entry updates to only include reasoning messages, simplifying the logic and improving clarity.
- Added imports for various dashboard, metrics, and report tools to the analyst workflow.
- Updated the `runAnalystWorkflow` function to return a comprehensive output, including tool call results, charts created, and summary statistics.
- Implemented logic to extract tool calls and segment messages by user requests, improving the overall workflow output structure.
- Upgraded '@aws-sdk/client-s3' to version 3.873.0 across multiple packages.
- Introduced caching mechanisms for metric data retrieval in the getMetricDataHandler function.
- Updated API endpoints to support report_file_id for cache lookups and data retrieval.
- Enhanced error handling and logging for cache operations.
- Refactored related components to accommodate new caching logic and parameters.
- Simplified the `GeneratingContent` component to display a static loading message.
- Updated `ReportPageController` to determine if a report is being generated based on the current message context.
- Adjusted report status handling in the delta creation process to maintain 'loading' status during streaming.
- Enhanced logic in the modify reports helper to ensure file status reflects loading during streaming and only updates to completed or failed when all edits are finalized.
- Modified database query to preserve input order for new messages during updates.
- Introduced `executeMetricQuery` utility for standardized metric SQL query execution with retry logic.
- Updated `getMetricDataHandler` and metric tool execution functions to utilize the new query utility, improving error handling and result processing.
- Added metadata generation from query results to provide detailed insights into data structure.
- Refactored SQL validation to ensure only read-only queries are executed, enhancing data integrity.
Enhance SQL permission validation to include unauthorized columns in error messages. Update permission error message creation to handle both unauthorized tables and columns. Modify SQL execution functions to validate column permissions alongside table permissions, improving feedback for users lacking access.