Report Abuse

Basic Information

Autono is a ReAct-based autonomous agent framework for building robust single- and multi-agent systems that dynamically generate next actions during execution. It implements an adaptive execution model that produces actions based on prior trajectories rather than fixed planner workflows. The repository provides a Python API to instantiate agents, declare abilities via decorators, customize personalities, and modify abilities at runtime. It includes a timely abandonment strategy using probabilistic penalties to manage termination behavior and a memory transfer mechanism to enable shared, dynamically updated memory among collaborating agents. The design is modular and supports external tool integration, with explicit compatibility for the Model Context Protocol (MCP) to access tools over stdio, HTTP SSE, or WebSocket.

Links

Categorization

App Details

Features
ReAct-based adaptive decision making that generates next actions during runtime. Timely abandonment strategy that applies probabilistic penalties to balance conservative and exploratory execution. Memory transfer mechanism for shared and dynamically updated memory in multi-agent collaboration. Ability and agentic decorators to declare callable tools and agent-invoking abilities. Personality enumeration to tune agent behavior such as PRUDENT or INQUISITIVE. McpAgent and mcp_session utilities with StdioMcpConfig for MCP integration over stdio, HTTP SSE, and WebSocket. Hooks BeforeActionTaken and AfterActionTaken for observability and intervention. Simple Agent API for granting, depriving, assigning tasks, and executing with just_do_it. Installable via pip.
Use Cases
Autono improves robustness and adaptability in complex task execution by allowing agents to revise actions dynamically and to abandon unproductive paths probabilistically. The memory transfer mechanism and explicit multi-agent collaboration enable task decomposition and focused division of labor to increase efficiency and quality. MCP support lets agents call external tools and services hosted as stdio, SSE, or WebSocket MCP servers, enabling flexible action-space extension. Observability hooks allow developers to inspect and intervene in decision and execution steps, aiding debugging and monitoring. The provided quick start examples, API for abilities and personalities, and reproducible experiments illustrate practical usage and performance advantages over comparable frameworks in multi-step and failure-prone scenarios.

Please fill the required fields*