airefinery-sdk 1.21.0 release

Announcement: AI Refinery SDK Version 1.21.0 Released 🚀

We're excited to announce the release of AI Refinery SDK Version 1.21.0!
This version brings performance improvements based on the valuable feedback we've received from the community.

💥 What's New in Version 1.21.0

• New Trusted Agent Huddle agent:
  • CB Insights Agent: This is a new agent that integrates with CB Insights' market intelligence platform.

🚨 Upcoming Major Update: Default Model Switch

Starting December 1, the default model will change from meta-llama/Llama-3.1-70B-Instruct to openai/gpt-oss-120b.

What this means

• After Dec 1, applications without an explicit model override will automatically use openai/gpt-oss-120b.
• We expect minimal disruption, but performance and outputs may differ slightly.

How to prepare

1. Test with gpt-oss-120b before Dec 1
   To validate your workflows ahead of the switch, configure your application by updating the YAML config according to the following example:

   base_config:
     llm_config:
       model: "openai/gpt-oss-120b"
       
  orchestrator:
    agent_list:
      - agent_name: "Search Agent"

  utility_agents:
    - agent_class: SearchAgent
      agent_name: "Search Agent"

2. Continue using Llama 3.1 if needed
   To explicitly stay on meta-llama/Llama-3.1-70B-Instruct after Dec 1, set the base_config as the following example:

   base_config:
     llm_config:
       model: "meta-llama/Llama-3.1-70B-Instruct"
       
  orchestrator:
    agent_list:
      - agent_name: "Search Agent"

  utility_agents:
    - agent_class: SearchAgent
      agent_name: "Search Agent"

✅ Compatibility

• SDK versions ≤ 1.15.1 can still connect to the current Distiller API server without modification.
• Applications that have not yet migrated to API-Key–only or TokenProvider authentication must do so before upgrading.
  Refer to the Quickstart guide for details.

airefinery-sdk 1.20.0 release

Announcement: AI Refinery SDK Version 1.20.0 Released 🚀

We're excited to announce the release of AI Refinery SDK Version 1.20.0!
This version brings performance improvements based on the valuable feedback we've received from the community.

💥 What's New in Version 1.20.0

• New Config Validation API
  • This API enables you to validate a distiller configuration file before creating a project. This method helps catch configuration errors early in the development workflow by sending your configuration to the server for validation against the latest SDK without actually creating a project. For more details, please check the official documentation for this API.
• New Trusted Agent Huddle agent:
  • Pega Agent:
    This agent is designed to intelligently analyze business workflows in real time and generate context-aware answers using enterprise knowledge to help streamline issue resolution. It is based on the Pega platform.

🚨 Upcoming Major Update: Default Model Switch

Starting December 1, the default model will change from meta-llama/Llama-3.1-70B-Instruct to openai/gpt-oss-120b.

What this means

• After Dec 1, applications without an explicit model override will automatically use openai/gpt-oss-120b.
• We expect minimal disruption, but performance and outputs may differ slightly.

How to prepare

1. Test with gpt-oss-120b before Dec 1
   To validate your workflows ahead of the switch, configure your application by updating the YAML config according to the following example:

   base_config:
     llm_config:
       model: "openai/gpt-oss-120b"
       
  orchestrator:
    agent_list:
      - agent_name: "Search Agent"

  utility_agents:
    - agent_class: SearchAgent
      agent_name: "Search Agent"

2. Continue using Llama 3.1 if needed
   To explicitly stay on meta-llama/Llama-3.1-70B-Instruct after Dec 1, set the base_config as the following example:

   base_config:
     llm_config:
       model: "meta-llama/Llama-3.1-70B-Instruct"
       
  orchestrator:
    agent_list:
      - agent_name: "Search Agent"

  utility_agents:
    - agent_class: SearchAgent
      agent_name: "Search Agent"

✅ Compatibility

• SDK versions ≤ 1.15.1 can still connect to the current Distiller API server without modification.
• Applications that have not yet migrated to API-Key–only or TokenProvider authentication must do so before upgrading.
  Refer to the Quickstart guide for details.

airefinery-sdk 1.19.0 release

🚀 Announcement: AI Refinery SDK v1.19.0 Released ⚡️
We're pleased to announce the release of AI Refinery SDK Version 1.19.0, featuring new capabilities and bug fixes to enhance your development experience.

🛠 New Features & Improvements

• Model Fine-Tuning API: The API enables you to customize AI models with your personal data according to your specific requirements. The Fine-Tuning API allows you to:

  • Create custom models tailored to your specific use cases
  • Improve model performance on domain-specific tasks
  • Adapt pre-trained models in Hugging Face to your particular data distribution

• Moderations API: The moderations provided by AIRefinery and AsyncAIRefinery clients is designed to ensure safety and compliance in text-based content. It effectively analyzes input text to detect potentially harmful material across 13 critical categories, such as sexual content, harassment, hate speech, illicit activities, self-harm, and violence. By flagging such content, the API empowers users to take preemptive corrective actions, which could involve filtering inappropriate material or monitoring interactions to maintain a respectful environment. Please check our guide to learn more details of this new API.

• New Trusted Agent Huddle Agent:

  • Wolfram Agent: This is a third-party agent powered by Wolfram|Alpha and the Wolfram Language, designed to perform advanced computations, mathematical derivations, scientific data lookups, visualizations, and knowledge-based queries.

• Added support for system_prompt_suffix field in orchestrator YAML config to allow application-specific prompt customization.

orchestrator_config:
    system_prompt_suffix: <customized_suffix>

✅ Compatibility

• SDK versions ≤ 1.15.1 can still connect to the current Distiller API server without modification.
• Applications that have not yet migrated to API-Key–only or TokenProvider authentication must do so before upgrading. Refer to the Quickstart guide for details.

airefinery-sdk 1.18.3 release

🚀 Announcement: AI Refinery SDK v1.18.3 Released ⚡️

We're pleased to announce the release of AI Refinery SDK Version 1.18.3, featuring new capabilities and bug fixes to enhance your development experience.

🛠 New Features & Improvements

• Query transformation: The AnalyticsAgent can transform the original user query into one or more refined queries using query_transformation_examples and database or dataframe schema defined in the executor configuration. This provides more flexibility in customizing how the agent processes different types of user queries. Please check our Analytics Agent updated guide and tutorial for more details about the new feature.

✅ Compatibility

• SDK versions ≤ 1.15.1 can still connect to the current Distiller API server without modification.
• Applications that have not yet migrated to API-Key–only or TokenProvider authentication must do so before upgrading. Refer to the Quickstart guide for details.

airefinery-sdk 1.18.1 release

🚀 Announcement: AI Refinery SDK v1.18.1 Released ⚡️

We're pleased to announce the release of AI Refinery SDK Version 1.18.1, featuring new capabilities and important bug fixes to enhance your development experience.

🛠 New Features & Improvements

• Deep Research Agent
  The new Deep Research Agent performs structured, multi-step research, efficiently handling complex queries by breaking them into manageable components.

• New Trusted Agent Huddle Agent: ServiceNow Agent
  Introducing the ServiceNow Agent for workflow automation and intelligent support. It operates on the ServiceNow platform using the A2A protocol and supports both inbound and outbound calls. Learn more here.

• Table Description for Analytics Agent
  Enhanced the Analytics Agent with the ability to include optional table descriptions, enabling a clearer understanding of your data. See more details here.

• Token Validator Bugfix
  Addressed a key issue with token validation, ensuring more reliable authentication processes for SDK < 1.13.

• ASR & TTS Inference Bugfix
  Fixed issues in Automatic Speech Recognition and Text-to-Speech, enhancing the accuracy of speech processing.

✅ Compatibility

• SDK versions ≤ 1.15.1 can still connect to the current Distiller API server without modification.
• Applications that have not yet migrated to API-Key–only or TokenProvider authentication must do so before upgrading. Refer to the Quickstart guide for details:
  https://sdk.airefinery.accenture.com/setup/quickstart/
  Thank you for your continued feedback and support.

airefinery-sdk 1.17.1 release

🚀 Announcement: AI Refinery SDK v1.17.1 Released ⚡️

We are pleased to announce the availability of AI Refinery SDK v1.17.1. This release focuses on improving type-safety, error transparency, and authentication flexibility.

🛠 Improvements & Fixes

• Typed DistillerClient Responses
  • All public methods of DistillerClient now return fully typed objects for improved IDE auto-completion, static analysis (e.g., mypy), and overall developer experience.

• Enhanced Error Handling & Propagation
  • Server-side errors are now surfaced through a clearer exception hierarchy.

• air.auth.TokenProvider (OAuth 2.0 / OIDC Support)
  OAuth users can now authenticate with an access-token flow in addition to a static API key.

  from air.auth import TokenProvider
  from air import AsyncAIRefinery
  
  # Keycloak token-endpoint pattern:
  #   https://<kc-host>/realms/<realm>/protocol/openid-connect/token
  keycloak_token_url = (
      "https://keycloak.example.com/realms/my-realm/protocol/openid-connect/token"
  )
  
  token_provider = TokenProvider(
      url=keycloak_token_url,
      client_id="my-client",
      client_secret="s3cr3t",
      extra_fields={"audience": "my-api"},  # optional; adapt to your IdP
  )
  
  client = AsyncAIRefinery(api_key=token_provider)
  
  # ─────────────────────────────────────────────────────────────
  # Existing AI Refinery API-key authentication remains fully supported:
  # client = AsyncAIRefinery(api_key="your_static_api_key")

✅ Compatibility

• This is a non-breaking upgrade from version 1.16.0.
• SDK versions ≤ 1.15.1 can still connect to the current Distiller API server without modification.
• Applications that have not yet migrated to API-Key–only or TokenProvider authentication must do so before upgrading. Refer to the Quickstart guide for details:
  https://sdk.airefinery.accenture.com/setup/quickstart/

Thank you for your continued feedback and support.

airefinery-sdk 1.16.0 release

🚀 Announcement: AI Refinery SDK v1.16.0 Released ⚡️

We're excited to announce the release of AI Refinery SDK Version 1.16.0! This version brings performance improvements based on the valuable feedback we've received from the community.

🛠 Improvements & Fixes

• Client Update
  Removed air.auth–based session login flow. All requests now rely solely on API-key authentication. For more details, please check the Quickstart page of the official SDK documentation.

from air import AsyncAIRefinery

client = AsyncAIRefinery(api_key=API_KEY)

All the tutorials of the official SDK documentation have been updated accordingly.

• Human Agent
  Bug fixes and stability improvements for smoother flow execution.

• openai/gpt-oss-20b Support
  The SDK now supports the openai/gpt-oss-20b model, delivering a lower latency model performance.

• General Performance
  Broader performance tuning and bug fixes to improve overall throughput and developer experience.

✅ Compatibility

• SDK versions ≤ 1.15.1 can connect to the new Distiller API server without any changes.
• If your application code has not yet been updated to use the API-Key–only client authentication, you will need to update it when upgrading to version 1.16.0. Please check the SDK Quickstart page fore more information.

airefinery-sdk 1.15.1 release

Announcement: AI Refinery SDK Version 1.15.1 Released 🚀

We're excited to announce the release of AI Refinery SDK Version 1.15.1! This version brings performance improvements based on the valuable feedback we've received from the community.

💥 What's New in Version 1.15.1

• New Trusted Agent Huddle Agent: Now you can connect AI Refinery with Snowflake Cortex agents.

• General Performance: Broader performance tuning and bug fixes to improve overall throughput and developer experience.

✅ Compatibility

Fully compatible with SDK 1.5.0 and later.

airefinery-sdk 1.15.0 release

🚀 Announcement: AI Refinery SDK v1.15.0 Released ⚡️

We’re excited to announce the release of AI Refinery SDK v1.15.0—a stability-focused update that strengthens agentic flow execution, optimizes large-model performance, and broadens model support.

🛠 Improvements & Fixes

• Flow Superagent
  Bug fixes and stability improvements for smoother flow execution.

• Variable Memory Module
  Reliability enhancements and issue fixes to ensure consistent memory handling.

• openai/gpt-oss-120b Scaling
  Optimized scaling improvements to significantly reduce latency when working with this model.

• General Performance
  Broader performance tuning and bug fixes to improve overall throughput and developer experience.

✅ Compatibility

• Fully compatible with SDK 1.5.0 and later.

airefinery-sdk 1.14.2 release

🚀 Announcement: AI Refinery SDK v1.14.2 Released ⚡️

We’re excited to share the release of AI Refinery SDK v1.14.2—a focused update that expands model support, introduces greater flexibility for orchestration, and boosts stability under heavy workloads.

✨ What’s New

• Support for openai/gpt-oss-120b
  The SDK now supports the latest openai/gpt-oss-120b model, delivering enhanced scale and reasoning power for advanced use cases.

• Configurable Orchestrator Routing
  Task routing and hand-off via the Orchestrator are now optional.

  • New setting: enable_routing
  • When enabled: queries are decomposed and routed across agents.
  • When disabled: all queries flow directly to the first agent in agent_list.
    This gives you fine-grained control over orchestration complexity vs. direct handling.

🛠 Improvements & Fixes

• General performance tuning and reliability enhancements to boosts stability under heavy workloads.
• More polished error handling.

✅ Compatibility

• If upgrading from 1.13.x or earlier, please ensure you use the new API keys from the AI Refinery platform.
  You can now create your client with:

  from air import AsyncAIRefinery
  airefinery_client = AsyncAIRefinery(api_key=api_key)

• Fully compatible with SDK 1.5.0 and later.