Skip to main content
This page is actively maintained. Issues are resolved in priority order based on impact and user feedback.

​
Current Limitations

​
Platform Constraints

Global Knowledge Bases & Tools

Status: Removed in v0.2.1Global knowledge bases and tools have been removed to simplify configuration. All knowledge bases and tools must now be configured per agent.Migration: Move global configurations to individual agent definitions.
# ❌ Old approach (no longer supported)
global:
  knowledge_bases: [kb1, kb2]
  tools: [tool1, tool2]

# ✅ New approach
agents:
  - name: my_agent
    knowledge_bases: [kb1, kb2]
    tools: [tool1, tool2]
Status: Deprecated, removal planned for v0.3Manual RAG (Retrieval-Augmented Generation) configuration is being phased out in favor of UKnow integration.Action Required: Migrate to UKnow for knowledge base management before v0.3 release.Timeline: Manual RAG will be fully removed in v0.3 (December 2025)
Not all model providers support all features:
  • Tool Calling: Not supported by Magistral, devstral-medium, mistral-moderation, and mistral-saba models
  • Vision: Limited to specific models (see Model Selection)
  • Streaming: Supported by most models, but may have provider-specific limitations

​
Known Issues

​
High Priority

Boolean Edge Key Interpretation

Issue: Boolean edge keys yes and no are interpreted as True/False in YAML parsing.Impact: Conditional edges with boolean routing may fail or route incorrectly.Workaround: Use quoted keys in your YAML configuration:
edges:
  - from: decision_agent
    condition: "Should we proceed?"
    condition_type: boolean
    routing:
      "yes": approval_agent  # ✅ Quoted
      "no": rejection_agent  # ✅ Quoted
Status: Fix planned for v0.2.2
Issue: The first user message in AMAS Chat appears after the first agent message instead of before.Impact: Chat history display order is incorrect for the initial exchange.Workaround: None currently. This is a UI display issue and doesn’t affect functionality.Status: Fix in progress

​
Medium Priority

Issue: OneDrive configuration also queries from SharePoint sites when using broad drive IDs.Impact: May retrieve documents from unintended SharePoint locations.Workaround: Only include drive IDs ending with .../me/<your_email> to limit scope to personal OneDrive:
knowledge_base:
  provider: uknow
  config:
    sources:
      - type: onedrive
        drive_id: "https://graph.microsoft.com/v1.0/me/[email protected]"
Status: Investigating improved filtering options
Issue: Switching browser windows or tabs may cause loss of selected agent context.Impact: Users need to reselect their agent after switching windows.Status: Fixed in v0.2.1
This issue has been resolved. Update to v0.2.1 or later.
Issue: Authorized UI doesn’t update immediately after successful authorization.Impact: Users may not see authorization status change without page refresh.Status: Fixed in v0.2.1
This issue has been resolved. Update to v0.2.1 or later.

​
Low Priority

Issue: Google Drive provider is marked as “insecure” in UKnow integration.Impact: Google Drive integration is available but flagged with security warning.Reason: UKnow is currently undergoing security re-issuance for Google Drive OAuth.Status: Official support coming shortly after security certification
Issue: Pydantic validation error messages could be more user-friendly.Impact: Technical error messages may be difficult for non-developers to understand.Status: Improved in v0.2.1
Error messages have been enhanced in v0.2.1 with clearer, more actionable feedback.

​
Feature Limitations

​
Agent Configuration

Current Limitations:
  • Maximum of 50 agents per workflow
  • Maximum of 100 edges per workflow
  • Circular dependencies are not automatically detected (will cause runtime errors)
  • Parallel execution limited to 10 concurrent agents
Planned Improvements (v0.3):
  • Automatic circular dependency detection
  • Increased parallel execution limits
  • Workflow validation before deployment

​
Performance Considerations

​
Response Times

Typical Response Times

  • Simple LLM Agent: 1-3 seconds
  • React Agent with Tools: 3-10 seconds
  • Multi-Agent Workflow: 5-30 seconds
  • Vision/OCR Tasks: 5-15 seconds

Factors Affecting Speed

  • Model selection (GPT-4o-mini is faster)
  • Number of tools available
  • Knowledge base size
  • Workflow complexity
  • Image/document size
For latency-sensitive applications, use GPT-4o-mini or Mistral Medium models and limit the number of tools per agent.

​
Token Usage

Context Window Management: Large knowledge bases and long conversation histories can quickly consume token limits.
Best Practices:
  • Use include_history: true with history_length: 5 instead of include_full_history: true
  • Implement conversation summarization for long sessions
  • Limit knowledge base scope to relevant documents
  • Use structured outputs to reduce response verbosity

​
Reporting Issues

Check This Page

Verify if your issue is already documented with a workaround

Search Documentation

Review relevant documentation sections for configuration guidance

Gather Information

Collect error messages, YAML configuration, and steps to reproduce

Contact Support

Email [email protected] with:
  • Clear description of the issue
  • Expected vs actual behavior
  • Configuration snippets (remove sensitive data)
  • Screenshots if applicable

​
Issue Tracking

Report Bug

Subject: “Bug: [Brief Description]”

Request Feature

Subject: “Feature Request: [Your Idea]”

Get Help

Subject: “Support: [Your Question]“

​
Update Frequency

This page is updated with each release. Last updated: November 2025 (v0.2.1)

View Release Notes

See what’s been fixed and what’s new in the latest releases