Skip to main content
Try it now at atthene.ai or visit the documentation to learn more.

v0.2.2

Released: November 28, 2025

Bug Fixes

Agent Handoff

  • Fixed: Improved agent handoff reliability and stability
  • Fixed: Disabled syncing flag for UKnow to improve performance
  • Fixed: Only see your own databases for better data isolation
  • Fixed: Disabled Teams feature temporarily for stability improvements
  • Fixed: Agents sharing and Atthene usage tracking now working correctly
  • Fixed: Resolved RAG errors for more reliable retrieval
  • Fixed: PDF and image handling in Atthene AMAS now working properly
  • Improved: RAG Simple View for easier knowledge base management
  • Fixed: Right-click paste functionality in editor
  • Fixed: Add Tools feature now working correctly
  • Fixed: Failed messages are now properly saved

New Features

Full History Injection

Use {{ full_history }} to inject all conversation messages into prompts

START Entry Point

START is now a valid entry point for workflow configurations

Agent Statistics

View detailed agent statistics and access control information

Agent Ownership

View agent owners for better team collaboration and management

v0.2.1

Released: November 25, 2025

Bug Fixes

UI Authorization Updates

  • Fixed: Authorized UI not updating after successful authorization
  • Added: Reset button to YAML configuration editor
  • Improved: Frontend now properly updates authorization status and displays authorization pills
  • Fixed: Switching windows no longer causes loss of selected agent context
  • Improved: Better state management across window navigation
  • Improved: Pydantic validation error messages are now clearer and more actionable
  • Enhanced: Better error feedback for configuration issues
  • Fixed: Resolved issues with agent handoff reliability
  • Note: Improved structured output handling for handoff scenarios

New Features

Gemini Support

Added support for Google’s Gemini models

Configuration Cleanup

Removed global knowledge bases and tools for cleaner configuration

v0.2.0

Released: November 21, 2025 | Milestone: Major feature release
This is a huge milestone in shaping the flexible future of agentic systems. Big thanks to the team!

YAML Configuration Enhancements

Intelligent YAML Suggestions

New: Press Ctrl + Space in the YAML editor to show available syntax keys and configuration options.
This feature dramatically speeds up agent configuration by providing context-aware suggestions as you type.

Model Provider Integrations

Mistral AI

Full support for Mistral models

Telekom OTC Hub

Enterprise-grade model access

More Providers

Expanded model selection
Available Models:
agents:
  - name: my_agent
    agent_type: llm_agent
    llm_config:
      # Mistral AI models
      model: mistral-large-latest
      # model: mistral-medium-latest
      # model: pixtral-large-latest  # Vision support
      
      # Telekom OTC models
      # model: claude-sonnet-4
      # model: Qwen2.5-VL-72B-Instruct  # Vision support
      # model: gpt-oss-120b
      
      temperature: 0.7
Check out the model providers documentation to see all available models and their capabilities.

Bring Your Own Key (BYOK)

Cost Control

Want more control over your cost base and token usage? We now support API keys for:
  • Mistral AI
  • Telekom OTC Hub
How to use BYOK:
agents:
  - name: my_agent
    agent_type: llm_agent
    llm_config:
      model: mistral-large-latest
      api_key: your-mistral-api-key-here  # Override default key
      temperature: 0.7
BYOK gives you direct control over API usage, billing, and rate limits while using the Atthene platform. Perfect for multi-tenant systems or cost allocation per department.

UKnow Integration

Cloud Storage Connections

Connect your enterprise knowledge sources:
  • SharePoint: Access organizational documents
  • OneDrive: Personal and shared file access
  • Google Drive: Currently undergoing security re-certification
UKnow provides a unified interface for all your knowledge sources with:
  • Real-time synchronization
  • Permission-aware access
  • Automatic indexing
  • Smart retrieval
Known Issue: Google Drive provider is currently marked as “insecure” while UKnow undergoes security re-issuance. Official support coming shortly.

Conditional Edges

Advanced Workflow Logic

Level up your multi-agent experiences by embedding flow logic directly into edges.
You now have two edge types to choose from:
Direct connections between agents:
edges:
  # Simple static routing
  - from: START
    to: analyzer
  
  - from: analyzer
    to: processor
  
  - from: processor
    to: END
Complete Example:
name: Support Router
architecture: workflow

agents:
  - name: classifier
    agent_type: llm_agent
    prompt_config:
      system_prompt: Classify incoming support requests
  
  - name: technical_support
    agent_type: react_agent
    tools: [search, documentation]
  
  - name: billing_support
    agent_type: llm_agent

edges:
  - from: START
    to: classifier
  
  # Conditional routing based on classification
  - from: classifier
    condition: "Route based on: {{ classifier.output }}"
    condition_type: literal
    possible_outputs: [technical_support, billing_support]
Known Issue: Boolean edge keys yes and no are interpreted as True/False.Workaround: Use quoted keys in your YAML:
routing:
  "yes": agent_name
  "no": other_agent

Structured Output with BAML

Industry-Leading Reliability

Struggling with reliability in producing structured JSON output? We’ve got you covered.
Utilizing BAML (thanks to BoundaryML), we guarantee the highest reliability in the industry for structured outputs.
1

Define Your Schema

Specify the exact JSON structure you need
2

Guaranteed Validation

BAML ensures outputs match your schema every time
3

Type Safety

Get strongly-typed responses with full validation
4

Error Recovery

Automatic retry and correction for malformed outputs
Example Configuration:
agents:
  - name: data_extractor
    agent_type: llm_agent
    prompt_config:
      system_prompt: |
        Extract structured data from customer inquiries.
        Focus on identifying key information accurately.
    
    structured_output:
      enabled: true
      output_name: customer_data
      schema:
        customer_name:
          type: str
          description: Full name of the customer
        
        email:
          type: str
          description: Customer email address
        
        order_id:
          type: str
          description: Order or ticket ID if mentioned
        
        priority:
          type: str
          description: Priority level (high, medium, low)
        
        issue_category:
          type: str
          description: Category of the issue
Output Format:
{
  "customer_data": {
    "customer_name": "John Doe",
    "email": "[email protected]",
    "order_id": "ORD-12345",
    "priority": "high",
    "issue_category": "billing"
  }
}
Structured outputs are perfect for data extraction, classification, form filling, and any task requiring consistent JSON responses.

Enhanced Tooling

Microsoft Outlook

  • Send and receive emails
  • Manage calendar events
  • Access contacts
  • Create tasks

Microsoft Teams

  • Send messages to channels
  • Create and manage teams
  • Schedule meetings
  • Share files

OCR & Vision Capabilities

Multimodal Processing

Got images for your agentic system? No problem!
Using multi-modal models like GPT-4o and Mistral OCR, you can now:

Extract Text from Images

  • Scanned documents
  • Screenshots
  • Photos of text
  • Handwritten notes
  • Identify objects and scenes
  • Extract structured data from forms
  • Understand charts and diagrams
  • Process receipts and invoices
  • Combine text and image context
  • Answer questions about images
  • Generate descriptions
  • Compare visual elements

Enhanced Prompt Configuration

Fine-Grained Control

New configuration options for precise prompt engineering
Control whether user messages are included in the prompt:
agents:
  - name: "assistant"
    prompt_config:
      include_user_input: true

Known Issues

We’re actively working on fixes for the following issues. Workarounds are provided where available.
Issue: Boolean edge keys yes and no are interpreted as True/FalseWorkaround: Use quoted keys in your YAML configuration:
edges:
  - from: "decision_agent"
    to:
      "yes": "approval_agent"
      "no": "rejection_agent"
Issue: First message in AMAS Chat appears after the first agent messageStatus: Fix in progress
Issue: Google Drive provider is marked as “insecure”Reason: UKnow is currently undergoing security re-issuanceStatus: Will be officially supported shortly
Issue: OneDrive configuration also queries from SharePoint sitesWorkaround: Only include drive IDs ending with .../me/<your_email> to limit scope to personal OneDrive
knowledge_base:
  provider: "uknow"
  config:
    drive_id: "https://graph.microsoft.com/v1.0/me/[email protected]"

Get Started


Follow for Updates

Stay tuned for more AI content and further updates on the Atthene platform!