> ## Documentation Index > Fetch the complete documentation index at: https://docs.polpo.sh/llms.txt > Use this file to discover all available pages before exploring further. # Introduction > Composable cloud backend for agents with sandbox runtime, tasks, workflows, and Model Gateway # Polpo Polpo is an open-source agent runtime and a managed cloud platform. Define agents, teams, tools, memory, and workflows in a `.polpo/` directory, then run them locally or deploy the same configuration to Polpo Cloud. ## What you get * **OpenAI-compatible chat** — `POST /v1/chat/completions`, including streaming responses * **Configurable agents** — models, tools, identity, skills, shared memory, and agent-specific memory * **Tool execution** — a built-in tool catalog plus project-defined custom tools * **Isolated execution** — managed sandbox execution on Cloud and local process execution in the open-source runtime * **Model routing** — the managed Model Gateway, encrypted provider keys, or a custom OpenAI-compatible gateway * **Orchestration** — teams, tasks, missions, Agentic Loops, approvals, schedules, and playbooks * **Optional assessment** — expectations, metrics, reviews, and retry policies for work that needs validation ## Cloud and open source Polpo Cloud adds the managed control plane around the open-source runtime: * organization and project management * project API endpoints and API keys * a project database and sandbox pool * managed model access and usage metering * persistent project files and operational dashboards The open-source runtime can run without a Polpo account and supports file, SQLite, and PostgreSQL storage. It also includes reusable v2 runtime views and a single-tenant dashboard host. See [Local Development](/docs/platform/local-development) and [Self-hosting](/docs/platform/self-hosting). Create a project, deploy the generated agent, and call its API.