RobotPai - AI Agent Applications Tool

Overview

RobotPai is an open, developer-focused autonomous agent project published as a Hugging Face Space. It wires a state-graph orchestration (langgraph StateGraph / MessagesState) to pluggable LLM adapters and explicit tool bindings so a model can call search, retrieval, and utility tools as part of a conversational flow. The Space includes a Gradio-based evaluation runner that fetches questions from a scoring API, runs the agent, and can submit answers for evaluation — useful for rapid prototyping and hands-on experiments. ([huggingface.co](https://huggingface.co/spaces/ZeroTimo/RobotPai/blob/main/agent.py?utm_source=openai)) The implementation shows concrete integrations: WikipediaLoader, ArxivLoader, Tavily web search, a Supabase-backed vector store with HuggingFace embeddings for retrieval-augmented responses, and LLM bindings for Google Generative AI (Gemini), Groq, and Hugging Face endpoints (example: Meta-DeepLearning/llama-2-7b-chat-hf). Recent commits in April 2025 focused on system-prompt handling, adding a retriever tool, and updating LLM bindings, indicating active maintenance on prompt and retrieval workflows. The repo is intended as a demonstrator and development scaffold rather than a commercial turnkey product. ([huggingface.co](https://huggingface.co/spaces/ZeroTimo/RobotPai/blob/main/agent.py?utm_source=openai))

Key Features

• Pluggable LLM bindings: ChatGoogleGenerativeAI, ChatGroq, ChatHuggingFace adapters.
• Retrieval-augmented responses using Supabase vector store and HuggingFace embeddings.
• Built-in tool set: Wikipedia, ArXiv, Tavily web search, and basic math utilities.
• State-graph orchestration (langgraph StateGraph / MessagesState) for controlled flows.
• Gradio evaluation runner with HF login and submission to a scoring API for testing.

Example Usage

Example (python):

from agent import build_graph
from langchain_core.messages import HumanMessage

# Choose provider: 'google', 'groq', or 'huggingface' (as supported in the Space)
graph = build_graph(provider='google')

# Run a single conversational query through the compiled state graph
messages = [HumanMessage(content='Summarize the latest research on diffusion models in 3 sentences.')]
result = graph.invoke({"messages": messages})
answer = result['messages'][-1].content
print('Agent answer:', answer)

# Note: this example follows the Space's agent.py structure (tools, retriever, LLM bindings).
# Source: RobotPai agent implementation on Hugging Face Spaces. ([huggingface.co](https://huggingface.co/spaces/ZeroTimo/RobotPai/blob/main/agent.py?utm_source=openai))

Key Information

• Category: Agent Applications
• Type: AI Agent Applications Tool