Changelog

Changelog

Technical updates and improvements

Key updates:

2025-04-03

PDF Document Processing for Tasks

Key features: • Added support for processing PDF documents within Tasks • Enabled Tasks to handle both text and document-based inputs

2025-01-12

Flexible OAuth Client Configuration

What's new: - Added support for multiple grant types when creating OAuth clients:     Client credentials (for server-to-server integrations)     Refresh token (for long-running access)     Implicit (for browser-based applications)     Authorisation code (for web applications with backend) - Introduced the ability to set and update redirect URIs - Settings are persisted in the database

2024-11-28

Task Input Processors

What's new: - Introduced the concept of Input Processors to manipulate task inputs dynamically - Launched our first processor: URL Fetcher, which automatically retrieves content from web pages - Created an API endpoint to list all available input processors - Updated the task revision schema to support processor configuration

2025-04-22

2025-04-22

Enabled Gemini 2.5 Flash Preview

Rightbrain users can now access Google’s Gemini 2.5 Flash Preview model, offering: • Superior performance on complex reasoning tasks with hybrid reasoning capabilities • Enhanced coding capabilities for software development tasks • Better multimodal processing for text and images • Maintains fast response times while delivering higher quality outputs • Excellent price-to-performance ratio compared to other leading models

2025-04-16

2025-04-16

Added support for OpenAI o3 and o4-mini models

Expanded our model selection with OpenAI's latest reasoning-focused models, providing Rightbrain users with enhanced capabilities for complex Tasks: o3 • State-of-the-art reasoning • Superior performance on complex coding and mathematical tasks • Enhanced multimodal capabilities for image understanding • Strong instruction following with minimal guidance o4-mini • Exceptional performance-to-cost ratio for reasoning tasks • Strong mathematical and coding abilities despite smaller size • Fast response times with high-quality reasoning • Optimised for high-throughput applications • Impressive visual reasoning capabilities

2025-04-15

2025-04-15

Rolled out OpenAI GPT-4.1 model series

Here’s the improvements you can expect: 4.1 • More reliable instruction following for better Task output consistency • Extended context processing up to 1M tokens for handling large documents • Enhanced multi-hop reasoning capabilities for complex document analysis • Improved multimodal understanding for better image processing in Tasks • More cost-effective than previous generation models 4.1 mini • Delivers powerful performance with reduced response times • Better reasoning capabilities than many previous flagship models • Excellent balance of quality and cost for most production use cases 4.1 nano • Fastest response times in the GPT-4 family • Optimised for classification, extraction, and high-throughput use cases • Cost-effective option for latency-sensitive applications

2025-04-04

2025-04-04

Gemini 2.5 Pro Experimental is now live

Key features: • Added support for Google's Gemini 2.5 Pro Experimental model • Integrated advanced reasoning capabilities for complex problem-solving • Added support for improved multimodal processing across text, images, and data

Why you should try it: • Instant access to Google's most intelligent AI model with superior reasoning abilities • Improved performance on complex tasks requiring advanced problem-solving • Better handling of large datasets across text, code, image, and pdf documents • More accurate results on specialised technical and mathematical problems

2025-04-03

2025-04-03

PDF Document Processing for Tasks

Key features: • Added support for processing PDF documents within Tasks • Enabled Tasks to handle both text and document-based inputs

Benefits: • Expanded Task capabilities to work with PDF files directly • Improved document processing workflow for more versatile use cases • Simplified content extraction from PDF documents

2025-04-01

2025-04-01

Expanded Output Format Constraints

Key features: • Added 'enum' support for constraining values in output formats • Implemented array with items constrained to enum values • Added string with value constrained to specific options • Added support for top-level object type • Introduced 'object' as an alternative reference to 'nested_structure'

How it helps: • Greater control over output data structure and validation • More consistent data through predefined option sets • Improved flexibility in defining complex data structures • More intuitive naming conventions for developers • Better tracking of input preprocessing

2025-03-31

2025-03-31

URL Fetcher Fallback Support

Key features: • Added fallback text capability for the URL fetcher input processor • Implemented configurable fallback messaging when URL content is unavailable • Added comprehensive test coverage for the URLFetcher class

Benefits: • More reliable task execution when external URLs are unreachable • Graceful handling of network or access issues with customisable fallback content • Improved user experience with configurable error handling • Better resilience for tasks dependent on external web content

2025-03-14

2025-03-14

Multiple Image Support

What's new • Process multiple images within a single task • Handle comparative image analysis in one operation • Upload several images simultaneously

Why it matters • No need to create separate tasks for each image • Simpler workflows for image comparison use cases • More efficient handling of multiple visual inputs

2025-03-10

2025-03-10

New Model Release • Gemini 2.0 Flash and Gemini 2.0 Flash Lite added to the platform

Gemini 2.0 Flash • Next-generation architecture • Native tool integration • Full multimodal capabilities • 1 million token context window • Enhanced processing speed Gemini 2.0 Flash Lite • Optimized for efficiency • Low latency response times • Multimodal processing • Large context window • Reduced computational requirements

Benefits • Process and analyse significantly larger documents in a single session • Achieve faster response times for interactive applications • Reduce operational costs with the Lite version for appropriate workloads • Enable more complex reasoning with expanded context retention

2025-02-28

2025-02-28

Added Support for Claude Sonnet 3.7 and GPT-4.5 Preview

Claude Sonnet 3.7: - Hybrid reasoning mode that toggles between quick responses and "extended thinking" for complex problems - Better accuracy in instruction-following, coding, mathematics and physics- Large context window of 200K tokens for handling substantial documents GPT-4.5 Preview: - More natural conversational style with improved pattern recognition and emotional intelligence - Particularly strong at writing assistance, programming support and creative brainstorming - Advanced capabilities for tasks requiring sophisticated reasoning and creative output

Both models are available immediately when creating new tasks or updating task revisions, and they work seamlessly with all existing task configurations and input processors.

2025-02-27

2025-02-27

Parallel URL Fetching

What's changed: - Tasks now fetch multiple URLs in parallel using concurrent processing

Significantly faster execution times for tasks that retrieve data from several web sources.

2025-02-21

2025-02-21

Task Credit Usage Reports

What's new: - New API endpoint to view credit usage for specific tasks - Reports show total credits consumed and number of task runs - Credit usage is broken down by task revision for more precise tracking - Optional date filtering (up to three months of history)

Gives you clearer visibility into which tasks are using the most credits and how usage patterns change over time.

2025-02-18

2025-02-18

Improved Encryption Management

What's changed: - Enhanced encryption management protocols have been implemented to secure sensitive data and streamline encryption processes

Establishes a more robust foundation for handling sensitive data while protecting the privacy and security of Rightbrain and its users.

2025-02-18

2025-02-18

Per-Task Credit Tracking

What's new: - Each task run now records the exact credits charged - Credits are calculated based on input and output tokens - This information is stored directly with the task run data - Account credit deductions are more precisely linked to specific runs

Gives you clearer visibility into exactly how credits are being used across your tasks, enabling precise cost allocation and tracking over time.

2025-01-30

2025-01-30

Task Image Storage and Retrieval

What's new: - Images submitted with task runs are now automatically stored - Each task run tracks the paths to any associated files - New endpoint to retrieve images directly using a simple GET request - Files are served with the appropriate content-type headers

Makes it much easier to work with image-based tasks by providing persistent storage and simple access to original files.

2025-01-29

2025-01-29

Organisation Accounts and Model Management

What's changed: - Each organisation now has a dedicated account for tracking and managing credits - Models are now defined in the database rather than in code, providing more flexibility - Credit costs are calculated based on input and output tokens at task execution time - Breaking change: Tasks now use llm_model_id instead of the previous llm_model field

More accurate credit tracking with atomic updates and clearer visibility at the organisation level, plus the ability to update model costs without code changes.

2025-01-12

2025-01-12

Flexible OAuth Client Configuration

What's new: - Added support for multiple grant types when creating OAuth clients:     Client credentials (for server-to-server integrations)     Refresh token (for long-running access)     Implicit (for browser-based applications)     Authorisation code (for web applications with backend) - Introduced the ability to set and update redirect URIs - Settings are persisted in the database

With these options, you can now create OAuth clients that match your specific integration needs, simplifying management for various scenarios.

2025-01-10

2025-01-10

Improved Task Execution Timing

What's new: - Added LLM call timing data to task run outputs - Improved tracking of input processing durations - Now capturing original input values before processing for better comparisons - Refactored error handling for more precise performance measurements

Provides clearer visibility into which parts of your tasks take the most time, helping you optimize prompt design and input processing.

2025-01-02

2025-01-02

Field Exclusion for API Responses

What's new: - Added exclude_field[] query parameter across API endpoints - You can specify multiple fields to omit from responses (supports nested data structures using dot notation)

Gives you more control over the data you receive by reducing response size and focusing on what you actually need.

2024-12-23

2024-12-23

Public Tasks

What's new: - Tasks can now be set as 'public' or remain private (the default) - Public tasks can be executed by any authenticated user - No permission checks are needed to run public tasks - Task creators maintain full control over editing and management

Makes it easier to share useful tasks with the wider Rightbrain community and speeds prototyping and collaboration.

2024-12-23

2024-12-23

Updated to Langchain 0.3

What's changed: - Upgraded Langchain and related packages to the 0.3 series - Removed older compatibility code that's no longer needed - Updated HTTP request handling for better reliability - Adjusted several related dependencies to their optimal versions

Keeps our foundation current with the broader AI ecosystem, ensuring compatibility with the latest tools and techniques.

2024-12-11

2024-12-11

Added Support for Llama 3.3 70B

What's new: - Added the Llama 3.3 70B Versatile model through our Groq integration - Features a substantial 128K token context window - Knowledge cutoff extends to December 2023

Gives you access to one of the most capable open models available for complex reasoning, content generation, and handling long documents.

2024-12-10

2024-12-10

Flexible Tagging for Tasks and Revisions

What's new: - Create tags at either user or project level - Assign tags to tasks and individual task revisions - Filter tasks and revisions based on their tags - Manage tags through new API endpoints (create, update, list, fetch) - Support for hierarchical tag structures

Helps you organise your work more flexibly, group related tasks, create custom workflows, and quickly find specific revisions.

2024-12-06

2024-12-06

Assign Documents to Collections at Creation

What's new: - Added optional collection_id parameter when creating documents - Documents automatically have their embeddings added to the specified collection - Validation ensures the collection exists and proper edit permissions - No changes to existing flows if no collection is specified

Removes the need to separately add documents to collections after creation, improving organisation from the start.

2024-11-29

2024-11-29

Task Revision Annotations

What's new: - You can now add a text annotation (up to 255 characters) to any task revision - Annotations appear alongside other revision details in the task history - Can be added when creating new revisions or updated later - This field is optional and doesn't affect task functionality

Provides valuable context about why changes were made, helping document revision purposes and aiding team collaboration.

2024-11-29

2024-11-29

Task Timing Reports

What's new: - Task runs now record timing information for LLM call duration and input processor duration - New reporting endpoint provides average timings by task revision - Flexible time grouping options (hourly or daily) and date range filtering

Helps identify performance bottlenecks, understand the impact of prompt designs, and monitor changes in response times over time.

2024-11-28

2024-11-28

Task Input Processors

What's new: - Introduced the concept of Input Processors to manipulate task inputs dynamically - Launched our first processor: URL Fetcher, which automatically retrieves content from web pages - Created an API endpoint to list all available input processors - Updated the task revision schema to support processor configuration

Enables automated retrieval and injection of web content into tasks, making them more powerful and flexible.

2024-11-25

2024-11-25

Explicit Task Revision Activation

What's changed: - Task revisions now require explicit activation rather than automatically using the latest version - Added a new 'active revisions' concept to clearly indicate which version is in use - Supports multiple active revisions simultaneously (though weights are currently fixed) - Each task shows its active revisions in the task output

Gives you better control over your task lifecycle by allowing thorough testing before switching revisions and lays the groundwork for future weighted distribution.

2024-11-14

2024-11-14

Test Task Revisions

What's new: - Ability to create test revisions of tasks that remain separate from production - Test revisions are hidden from standard task lists by default - These revisions cannot be activated as the primary version - Can be executed explicitly for testing - Reported separately in usage and token reports

Provides a safe space to experiment with changes without affecting live operations, allowing iterative testing and feedback.

2024-11-07

2024-11-07

Better Ownership Controls for Project Clients

What's new: - Project-level OAuth client creator is automatically assigned as owner - All project members receive viewer access by default - Tasks created by service accounts grant editor access to all project members - Clearer ownership relationships between projects, clients, and tasks

Makes collaboration more intuitive by ensuring everyone on the project team has appropriate visibility and control over OAuth clients.

2024-11-06

2024-11-06

Refresh Tokens for User-Defined OAuth Clients

What's changed: - User-defined OAuth clients now support the refresh token grant type - The 'offline_access' scope is automatically included - No changes required for existing clients unless opting to use this feature

Lets your applications maintain continuous API access without requiring users to repeatedly authenticate.

2024-11-04

2024-11-04

Project-Level OAuth Client Permissions

What's new: - Added role-based permissions for OAuth clients (owner, editor, viewer) - Each role has specific capabilities (owners can manage members; editors can edit but not manage membership; viewers have read-only access) - Project members can be granted the ability to create OAuth clients

Enables sharing OAuth client access with team members while maintaining proper security boundaries and enforcing least privilege.

2024-10-15

2024-10-15

Added Client Context to /whoami Endpoint

What's new: - The /whoami response now includes client details such as Client ID, Organisation, and Project - This information is automatically derived from your authentication token

Provides more complete context about your API session in a single request, aiding in troubleshooting and permission verification.

2024-10-15

2024-10-15

Choose Between User and Organisation OAuth Clients

What's new: - When creating an OAuth client, you can now specify a type: either user (default) or organisation - User clients inherit the creator’s permissions; organisation clients use a service account with separate permissions

Offers more targeted access control, letting you choose between personal integrations and shared tools with consistent access.

2024-10-14

2024-10-14

Simplified API Permissions

What's changed: - Simplified task management by removing unnecessary scope requirements - Shifted to a more flexible permissions model based on Service Account authorisation - Streamlined token validation by removing redundant client ID checks

Makes API access more straightforward for service account integrations while maintaining security; existing tokens continue to work as before.

2024-10-11

2024-10-11

OAuth Client Management API

What's new: - Create your own OAuth clients programmatically through the API - List and fetch details of existing OAuth clients - Each client comes with an associated service account for permissions management - Add descriptions to help organise and identify your clients - Important: Client secret is only shown once at creation

Gives you more control over how you integrate with our platform by managing OAuth clients programmatically with enhanced security and organisation.

2024-10-03

2024-10-03

Smarter API Documentation Generation

What's changed: - Added automatic detection and population of missing path parameters in OpenAPI documentation - Created a custom hook that runs during documentation generation to fill in gaps - Fixed inconsistencies between actual API paths and their documentation

Results in more complete, reliable API documentation, making integration easier for developers.

2024-09-26

2024-09-26

Token Usage Tracking and Reporting

What's new: - Each task run now records input, output, and total token counts - Added a new reporting endpoint to analyse token usage across your tasks - Reports can be filtered by date ranges using natural language (e.g., "last 7 days") - Usage data can be broken down by individual task revisions

Helps you monitor costs and usage patterns, identify opportunities to optimise prompts, and make informed model selection decisions.

2024-09-09

2024-09-09

Run Specific Task Revisions

What's new: - When running a task, you can now specify a particular revision to use - If no revision is specified, the system defaults to the active revision - Added validation to ensure the requested revision exists and belongs to the task

Provides more control over task execution, valuable for testing new revisions and ensuring consistent results even if the active revision changes.

2024-09-06

2024-09-06

Groq Integration with Llama 3.1 70B and Mixtral 8x7B Models

What's new: - Integrated Groq as a model provider- Added Llama 3.1 70B and Mixtral 8x7B models - Improved task prompts to ensure valid JSON responses - Better handling of invalid responses from language models

Offers models with larger context windows and improved performance for complex tasks, balancing advanced reasoning and efficiency.

2024-09-05

2024-09-05

New Task OAuth Client Management Tool

What's new: - Created a CLI script to quickly set up OAuth clients for task execution - Added validation to check project and user IDs before creating clients - Improved logging with better object representation for debugging- Updated dependencies to maintain security and compatibility

Simplifies the process of setting up secure authentication for tasks, ensuring proper configuration without duplicate clients.

2024-09-04

2024-09-04

Feature Flag Management System

What's new: - Introduced a feature gating mechanism to toggle features on or off based on specific criteria - Added controls to target features by user, organisation, and project - Created a new service to manage feature flags

Enables smoother feature rollouts with gradual, controlled access and better risk management through quick disablement if issues arise.

2024-08-29

2024-08-29

New Cohere Models Now Available

What's new: - Added support for Cohere's command, command-r, and command-r-plus models - Improved message handling for models that don't support multi-modal inputs - Standardised messages into a consistent format across all models

Provides longer context windows, higher token output limits, and broader model selection to choose the right tool for your specific use case.

2024-08-21

2024-08-21

Background Tasks Reorganisation

What's changed: - Moved several task-related imports to a new location to keep the code more organised - Created a new utilities module for task prompt validation functions - Separated RAG configuration validation into dedicated helper functions

Improves code organisation and maintainability without changing functionality, aiding future development.

2024-08-20

2024-08-20

Improved Validation for Task Prompt Variables

What's new: - Variables in prompts must now be properly enclosed with curly braces {variable} - Variable names must consist of alphabetic characters only - Clear feedback provided when invalid variables are detected

Helps catch common formatting mistakes early in the task creation process, providing immediate and specific guidance on corrections.

2024-08-05

2024-08-05

Better Handling of Non-JSON Responses from LLMs

What's changed: - Non-JSON responses from language models are now properly identified as errors - Returns a clear 500 status code with detailed information - The problematic response is captured in the error message for easier debugging - Failed LLM responses no longer trigger response forwarding

Provides clearer error feedback and more robust handling of unexpected LLM outputs, reducing noise in webhook endpoints.

2024-07-25

2024-07-25

Complete Task Run Data in Webhook Payloads

What's improved: - Webhook payloads now include the full task run output including metadata, timestamps, and submission details - Eliminates the need for additional API calls to retrieve complete run information - Better support for audit trails and system integrations with richer data

Gives you immediate access to all task run details directly in your webhook handlers, simplifying integration and auditing.

2024-07-23

2024-07-23

Added Task ID to Run Outputs

What's new: - Added the parent task_id field to all task run output responses

Simplifies reconciliation between tasks and their individual runs, improving traceability in reporting and monitoring.

2024-07-22

2024-07-22

Improved Task Error Handling

What's changed: - Standardised error formats across all Task operations - Moved input validation earlier to catch errors before task execution - Reduced unnecessary processing of invalid tasks, improving response times and reducing costs

Provides more consistent and helpful feedback when errors occur, making troubleshooting easier and more efficient.

2024-07-11

2024-07-11

API Usage Metering

What's new: - Added measurement of API endpoint calls by resource type and operation - Implemented daily tracking intervals for precise usage visibility - Started with Task endpoints, with plans to expand

Helps you understand usage patterns and supports future reporting capabilities by collecting valuable usage data.

2024-06-24

2024-06-24

Free Tier Registration & JWT Authentication Support

What's new: - Free Tier Support: Users without an active invitation can now register and gain immediate access to a free tier project. JWT Authentication Support: Added support for user creation via Hydra-issued JWT access tokens, enabling more flexible authentication workflows

Makes the platform more accessible for new users and provides additional secure authentication options for diverse integration scenarios.

2024-06-24

2024-06-24

Task Revisions

What's new: - Reworks the underlying schema for Tasks to introduce the concept of Task Revisions - Enables future functionality like A/B testing prompts, model performance evaluation, and model vendor load balancing

Establishes a foundation for advanced revision management, giving you better control over task versions and future functionality enhancements.

2024-05-24

2024-05-24

Enable RAG on Tasks

What's new: - Upgrades Tasks to allow for the configuration of RAG against existing RB collections - Enables more detailed output format definition

Improves task configuration for enhanced information retrieval and custom output formatting.

2024-05-24

2024-05-24

Enable VertexAI powered Gemini models

What's new: - Added support for Gemini models

Expands the range of available models, offering more options for diverse AI task requirements.

2024-04-24

2024-04-24

OAuthClient support in API

What's new: - Adds functionality to create an OAuth client application in Hydra, with a DB entry linking the creating author, project, and org

Automates and streamlines OAuth client registration and association with projects and organisations.

2024-04-24

2024-04-24

Enable Anthropic and conditional LLM interfaces

What's new: - Added support for Anthropic models - New model_factory uses type definitions to determine the LLM interface to return

Broadens model support and allows conditional interfacing based on model type, enhancing flexibility in LLM integrations.

2024-02-24

2024-02-24

Framework for Conversational Guardrails

Key Features: - Configurable input guardrails at both project and chat levels - Two initial guardrail policies: prompt policy and PII policy - Seamless integration with existing chat configurations - Compatible interface design for future extensions

Helps organisations maintain control over AI interactions by filtering inappropriate content and protecting sensitive information, aligning with security and compliance requirements.

2024-02-24

2024-02-24

Audit Logging

Key Features: - Dedicated, secure storage for all project-related activity - Encrypted audit logs using customer-managed keys - Standardised logging format for consistent record-keeping

Benefits: Provides critical visibility into system interactions, helping organisations meet compliance requirements and maintain proper governance over AI operations.

Join our developer slack

Request to join our developer slack channel

Join us on

Join our developer slack

Request to join our developer slack channel

Join us on

Join our developer slack

Request to join our developer slack channel

Join us on

Join our developer slack

Request to join our developer slack channel

Join us on