examples

Full list of examples of how to use the TapeAgents framework for building, debugging and improving agents. Each example is a self-contained Python script that demonstrates how to use the framework to build an agent for a specific task.

Examples

In the order of increasing complexity:

• llama_agent.py - simplest agent that uses LLaMA model to answer to user in a style of Shakespeare.
• llama_user.py - conversation between the LLaMA agent and the agent that emulates the user behavior.
• continue_tapes.py - agent that continues the tape on behalf of different user.
• batch_main_loop.py - batch processing of the tapes.
• batch_add_observations.py - batch processing when emulating multiple users.
• chat.py - demo of two agents chatting with each other.
• openai_function_calling.py - agent that uses OpenAI API function calling to report weather in a city.
• openai_function_calling_demo.py - interactice Gradio demo of the previous agent.
• tools_demo.py - demo of the previous agent with external web search tool.
• agent.py - examples of using nodes to control the agent.
• code_chat.py - simple agent that can solve tasks using python code.
• annotator.py - example of the agent that annotates the existing tape with some score or label.
• annotator_demo.py - interactive Gradio demo of the previous agent.
• multi_chat.py - multi-agent setup where team of the agents collaborates to answer to user. Includes group manager agent, software developer agent and code executor agent.
• data_science.py - data-science oriented multi-agent setup that solve a single data processing task using python.
• workarena - custom agent that solves WorkArena benchmark using BrowserGym environment.
• gaia_agent - custom agent that solves Gaia benchmark using planning and a set of tools with web search, documents and media parsers, code execution.
• delegate.py - multi-agent setup where agent contains multiple sub-agents and delegates the tasks to them.
• delegate_stack.py - more complex multi-agent setup where agent uses stack of the tape views to track data for each sub-agent.
• studio.py - interactive Gradio demo of agent that could be edited in runtime.
• tape_improver - the agent that revisit and improves the tapes produced by another agent.
• gsm8k_tuning - custom agent that solves GSM-8k benchmark, collect tapes and finetune smaller LLaMA model on them.