Super Agent Usage

Setup

1. Install uv if it is not already on your PATH.
2. From the repo root run uv sync to create/update the virtual environment with the pinned dependencies in uv.lock.
3. Run uv pip install tiktoken separately
4. Configure your LLM credentials in a .env script:
   • API_BASE
   • API_KEY
   • MODEL_NAME
   • MODEL_PROVIDER

Usage

• Run the primary Super Agent demo (covers tool usage, context management, streaming, token accounting, and sub-agent delegation):

  uv run python examples/super_agent/test/super_react_agent_example.py
  

  The script prints the progress of each scenario and exits once all seven demonstrations complete.

• Run the Super Agent demo with MCP coverage:

  uv run python examples/super_agent/test/super_react_agent_example_mcp.py
  

  Do note that you will need MCP servers to run in order to run this example. The MCP servers need to be SSE and not stdio. Optionally, feed the output into a .txt file. e.g. uv run python examples/super_agent/test/super_react_agent_example_mcp.py >> example_result.txt

• In order to run batches of datapoints with MCP (e.g. for GAIA dataset):

1. Provide a JSONL task file to your own harness (see examples/super_agent/data/test.jsonl)
2. Run the Super Agent test run script that allows for running multiple questions.

uv run python examples/super_agent/test/super_react_agent_test_run.py

Optionally, feed the output into a .txt file. e.g. uv run python examples/super_agent/test/super_react_agent_example_mcp.py >> example_result.txt