Polpo is an AI agent runtime that acts as an infrastructure layer for building and deploying production-grade agents. It allows developers to create agents for both conversational and long-running tasks that can remember past actions, use tools, and execute real work inside isolated sandboxes.

Instead of treating agents as simple chat interfaces, Polpo positions them as autonomous workers capable of performing side-effect-heavy tasks like coding, file manipulation, API calls, and workflow automation. These agents can operate continuously or on demand, and are designed to function reliably at scale.

The platform exposes a single, OpenAI-compatible API that works with any model provider. Developers define agents using a declarative configuration (typically JSON), specifying roles, tools, skills, memory access, and execution constraints. Once deployed, agents are immediately accessible via HTTP endpoints.

A core part of Polpo is its sandboxed execution environment, where every agent run is isolated with its own filesystem and tool access. This enables safe execution of real-world actions such as running shell commands, editing files, or interacting with external services without risking system-wide impact.

Polpo also includes persistent memory, tool orchestration, scheduling for long-running tasks, and multi-agent coordination. Agents can spawn other agents, delegate subtasks, and operate as structured teams rather than isolated prompts.

The system is designed for production use cases where reliability matters: support automation, autonomous workflows, coding agents, and background task execution. It supports both cloud-hosted and self-hosted deployments and is framework-agnostic.

Key features include:

• Unified OpenAI-compatible API for all agents
• Declarative agent definitions (JSON-based)
• Persistent memory across sessions
• Tool system (files, shell, HTTP, search, etc.)
• Sandboxed execution environments per run
• Long-running and scheduled agent tasks
• Multi-agent orchestration and delegation
• Skills system for reusable capabilities
• Support for any LLM provider
• Horizontal scaling (single or thousands of agents)

Common use cases include:

• Building autonomous AI applications
• Coding and dev-automation agents
• Customer support and operations agents
• Multi-step workflow automation
• Background task processing and scheduling
• AI systems with persistent memory and tool use
• Agent-based SaaS backends

Polpo is developed by Alessio Micali (founder of Lumea Technologies) as an open-source agent runtime focused on making AI agents production-ready and infrastructure-native.