State1 0.6.7

A unified framework for building AI agents with low-code.

State-1: Unified AI Agent Framework

State-1 is an open-source Python framework for building powerful AI agents with a unified, low-code experience. It abstracts away the complexity of LLMs, web search, RAG, CoT, UI, actions, authentication, memory, and multi-agent orchestration—so you can build, extend, and run agents with a single package and simple commands.

---

Features

• Unified Agent API: One class, many capabilities—LLM, RAG, web search, CoT, actions, UI, and more.
• LLM Integration: OpenAI and OpenRouter support, with function calling and model selection.
• Web Search & RAG: DuckDuckGo search and document ingestion (txt, PDF, DOCX) with vector search.
• Chain-of-Thought (CoT): Step-by-step reasoning for more robust answers.
• Actions: Real-world actions (send email, fetch API) with OpenAI function calling.
• Authentication & Memory: User login/signup, session management, and persistent chat memory.
• Multi-Agent Orchestration: Parallel, sequential, voting, and manager/worker workflows.
• Extensible: Add your own actions, agents, and workflows easily.

---

Installation

$ pip install state1 .

---

Quick Start: Terminal Chat

Run the terminal chat interface:

$ terminal-chat

On first run, the following demo agents are auto-created and ready to use:

1. ActionDemo Agent

• Description: Can send emails and fetch data from APIs using natural language.
• Actions: SendEmailAction, FetchAPIAction (pre-registered)
• Example: "Send an email to [e-mail] with subject Hello and body Hi Alice" or "What is the weather in London?"
• Setup: Edit the SMTP and API config in state1/terminal_chat.py or via the agent file in agents/.

2. RAG Agent

• Description: Retrieval-Augmented Generation agent. Ingests documents (txt, PDF, DOCX) and answers questions using them.
• Usage: Add documents with agent.add_document('path/to/file') in Python, or extend the agent in code.
• Example: "What does the example.pdf say about climate change?"

3. AuthTest Agent

• Description: Requires user login/signup (email & password). Remembers chat history per user.
• Usage: On first chat, you'll be prompted to log in or sign up.

4. Orchestrator Demo

• Description: Multi-agent orchestration (manager/worker workflow). Collaborates with Researcher, Summarizer, FactChecker, Analyser, and Manager agents.
• Usage: Switch to this agent and ask any question to see multi-agent collaboration.

---

First Run Experience

When you run terminal-chat for the first time after installation, the following showcase agents are automatically created:

• Default Agent: Minimal, just LLM chat.
• Web Search Agent: Only web search enabled.
• RAG Agent: Only RAG enabled (add documents via Python or code).
• CoT Agent: Only Chain-of-Thought enabled.
• Actions Agent: Only actions enabled, with example SMTP/API config and actions registered.
• Auth Demo Agent: Only auth and memory enabled.

Switch between agents with /list and /switch <agent_id>. Each agent demonstrates a single feature for easy testing and learning.

---

Test Checklist: All Features

After installing and running terminal-chat, test the following:

1. Default Agent

   • Chat with the agent. It should behave as a basic LLM assistant.

2. Web Search Agent

   • Switch to the Web Search Agent.
   • Ask a question that requires web search (e.g., "What's the latest news?").
   • Confirm the answer includes web search results.

3. RAG Agent

   • Switch to the RAG Agent.
   • In Python, use agent.add_document('path/to/file') to add a document.
   • Ask a question about the document and confirm the answer uses document content.

4. CoT Agent

   • Switch to the CoT Agent.
   • Ask a complex question and confirm the agent reasons step by step.

5. Actions Agent

   • Switch to the Actions Agent.
   • Ask to send an email or fetch API data (edit SMTP/API config as needed).
   • Confirm the agent performs the action and returns a result.

6. Auth Demo Agent

   • Switch to the Auth Demo Agent.
   • You should be prompted to log in or sign up.
   • Confirm chat history is saved per user.

7. Create a Custom Agent

   • Use /new and enable any combination of features.
   • Confirm the new agent works as expected.

8. Orchestrator Demo

   • Use /orchestrator to launch the multi-agent orchestrator.
   • Try different workflow modes (parallel, sequential, voting, manager).

---

If any feature does not work as expected, check your API keys, SMTP config, and document paths. See the agent's description for its enabled features.

---

Switching Between Agents

• List all agents:

  /list
  

• Switch to an agent:

  /switch <agent_id>
  

  (You can use the first few characters of the agent ID.)

---

Creating & Extending Agents

• Create a new agent interactively:

  /new
  

• Extend agents in Python:
  • Add actions: agent.register_action(MyCustomAction())
  • Ingest documents: agent.add_document('myfile.pdf')
  • Add new workflows or orchestration: see state1/orchestrator.py

---

Configuration

• API Keys: Replace YOUR_OPENAI_API_KEY in the agent configs or set the OPENAI_API_KEY environment variable.
• SMTP/Email: Use an app password and correct SMTP server/port for email actions.
• APIs: Add or edit API configs in the agent's apis dictionary.

---

Advanced: Multi-Agent Orchestration

• The Orchestrator agent supports parallel, sequential, voting, and manager workflows.
• To try different workflows, edit the orchestrator agent config in agents/orchestrator-demo.json or in code.

---

Contributing

Pull requests, issues, and feature suggestions are welcome! See CONTRIBUTING.md (if available) or open an issue on GitHub.

---

License

MIT License. See LICENSE.