v1.7.0

[1.7.0] - 2025-03-23

Added

• OpenAI Computer Use Integration:

  • Integrated a new tool for browser automation, OpenAI Computer Use, which mirrors the capabilities of the OpenAI Operator. This tool is designed to execute advanced browser automation tasks with enhanced efficiency. You can find the new tool in default_tools.py under the name browser_navigation_cua.

• WebSocket Integration:

  • Added a WebSocket module to provide real-time updates to the frontend. This integration continuously streams all reasoning steps and backend operations, improving transparency and user interaction.

• OpenAI Web Search Tool Integration:

  • Integrated a new web search tool that utilizes the Chat Completions API. With this integration, the model retrieves information from the web before responding to queries, leveraging fine-tuned models and tools similar to those used in Search in ChatGPT. You can find the new tool in default_tools.py under the name helper_model_web_search.

v1.4.0

[1.4.0] - 2025-02-22

Added

• Deep Search Integration:
  • Introduced a multi-level deep search mode that adjusts key parameters (token count, scrape length, and search results) based on the specified depth.
• EGOT System:
  • Implemented the Evolving Graph of Thought (EGOT) framework to dynamically map and visualize the agent’s reasoning process using Neo4j.
• Persistent Memory with Redis:
  • Added persistent session management using Redis, ensuring that session data and memory logs are maintained across interactions with the same user.
    

v1.2.0

[1.2.0] - 2025-02-11

Added

• Function Validator:
  • Introduced a comprehensive function validator that leverages Python's ast module to analyze and verify function code before execution.
  • The validator performs multiple checks including:
    • Syntax Validation: Ensures the code is syntactically correct.
    • Import Restrictions: Verifies that only allowed libraries are imported and disallows relative (local) imports.
    • Parameter Signature Checks: Validates that function definitions adhere to expected parameter rules (e.g., no varargs, proper defaults for keyword-only arguments, and correct naming for subtask-specific parameters).
    • Assignment Verification: Confirms the presence of mandatory assignments like updated_dict = previous_output.copy() for subtasks with index > 0.
    • Undefined Name Detection: Identifies any use of undefined names within the function.
    • Nesting Depth Control: Ensures function nesting does not exceed one level (primary function plus optional helper functions).
    • Dangerous Function Call Prevention: Blocks execution of dangerous functions (e.g., eval, exec, compile, __import__, os.system, etc.).
    • updated_dict.get Key Validation: Checks that keys used in updated_dict.get(...) calls exist in the provided previous_output.
  • Upon successful validation, the function is automatically renamed to match the expected subtask name.
• RAG Llama Index Integration:
  • Added capabilities for llama index retrieval and ingest, expanding the system's data ingestion and search functionality.
• Subtask Regeneration Function:
  • Implemented a new function that automatically regenerates a subtask if:
    • The function validator fails during pre-execution checks, or
    • An error is encountered during execution.