What is Tinymanus

TinyManus is a lightweight, pluggable multi-agent runtime designed for LLM-based task collaboration. It offers a clean agent abstraction, task-level flow control, and session-based context sharing.

Use cases

Use cases for TinyManus include multi-agent systems for collaborative problem solving, automated content generation, real-time data processing, and educational tools that require agent collaboration.

How to use

To use TinyManus, define agents implementing the run method, create a session with a task prompt template, and set the task flow for execution. The session can then be run to manage the collaboration between agents.

Key features

Key features include a simple agent interface, task-centric workflows, hybrid execution plans, optional planner for auto-generating task flows, dynamic prompt injection with context, and enhanced logging for real-time monitoring.

Where to use

TinyManus can be used in various fields such as artificial intelligence, collaborative robotics, automated customer service, and any domain requiring multi-agent task orchestration.

🧠 TinyManus: A Lightweight Multi-Agent Collaboration Framework

TinyManus is a minimal, extensible multi-agent orchestration framework designed for LLM-based task collaboration. It provides a clean agent abstraction, task-level flow control, and session-based context sharing, with support for both user-specified workflows and automated planning.

🚀 Features

• ✅ Simple Agent Interface
  Agents are decoupled from orchestration logic — just implement run(request: str) -> str.

• 🧩 Task-Centric Workflow
  Each task is bound to one agent and managed through a Session-level flow controller.

• 🔀 Hybrid Execution Plan
  Supports mixed sequential and parallel execution like [[Task1, Task2], [Task3]].

• 🧠 Planner Optional
  Use a PlannerAgent to auto-generate task flows, or manually define them for full control.

• 📝 Prompt Injection with Context
  Prompts are dynamically constructed using a template with prior task history.

• 📊 Logging & Monitoring
  Emoji-enhanced logging helps track execution in real time.

🧪 Minimal Example

agentA = ReActAgent("AgentA")
agentB = ReActAgent("AgentB")
agentC = ReActAgent("AgentC")

template = TaskPromptTemplate("""
You are a member of a multi-agent collaboration system.
Your current task:
----
{task}
----
Previous task history:
----
{history}
----
""")

session = Session("TinyDemo", template)

session.set_task_flow([
    [Task("Introduce TinyManus", agentA), Task("Give a use case", agentB)],
    [Task("Summarize the above", agentC)]
])

await session.run()

📁 Project Structure

TinyManus/
├── agent/                  # Core agent definitions
│   ├── base.py             # BaseAgent interface
│   ├── one_shot.py         # One-shot agent variant
│   └── react.py            # ReAct-based agent logic
├── memory/                 # Memory backend
│   └── memory.py
├── prompt/                 # Prompt templates
│   ├── react_prompt.py
│   └── template.py         # Shared prompt template
├── test/                   # (Reserved for unit tests)
├── tools/                  # Built-in tools
│   ├── base.py
│   ├── google_search.py
│   ├── terminator.py
│   ├── tool_collection.py  # Tool registry
│   └── cmd/                # Shell-based tools
│       ├── shell_tool.py
│       └── shell_session.py
├── utils/
│   └── logger.py           # Logging utilities
├── workflow/               # Task/session orchestration
│   ├── session.py
│   └── task.py
├── llm.py                  # LLM interfaces or wrappers
├── main.py                 # Entry point
├── LICENSE
└── README.md

🔮 Coming Soon

• [ ] Task timeout / retry control
• [ ] DAG-based execution planning
• [ ] Visual session dashboard
• [ ] Toolchain system enhancements
• [ ] MCP (Model Context Protocol) support
• [ ] Multi-agent debate mode with training data generation
• [ ] Long-term Memory
• [ ] And More…

🤝 Contributing

Pull requests are welcome! For major changes, please open an issue first to discuss what you would like to change or improve.

🧑‍💻 Author

Created by Zhiyu Zhang, UC Davis.

🪄 License

Apache 2.0 License. See LICENSE for details.