Hugin

A framework for building agents with a focus on longer running, creative, reasoning tasks.

Every agent step is a state on a stack. Easily replay, step-through, branch, and debug your agent's reasoning.

Built-in support for parallel agents, shared state with namespaces, and both synchronous and asynchronous agent-to-agent communication.

Real-time monitoring of agent flows, tool calls, and decision trees. Understand exactly what your agent is doing.

YAML-based configs for agents, tasks, and tools. Python for custom tool implementations.

Built-in support for roll-outs and parallel reasoning using branching of reasoning traces.

Built-in AskHuman interactions for approval workflows, human oversight, and interactive agent guidance.

Long-term memory through artifacts. Agents can save insights, documents, and data that persist across sessions.

Full execution history saved to disk. Replay from any state, resume interrupted runs, inspect past decisions.

Built-in tools, multiple LLM providers (Anthropic, OpenAI, Ollama), and an interactive CLI to create agents in seconds.

Quick Example

pip install gimle-hugin
hugin create

The agent builder wizard will guide you through creating your first agent.

Or create one manually:

# configs/my_agent.yaml
name: my_agent
system_template: default
llm_model: haiku-latest
tools:
  - builtins.finish:finish

# tasks/my_task.yaml
name: my_task
prompt: "Analyze the data and provide insights."

uv run run-agent --task my_task --task-path ./my_agent