The Agent-Native Cloud

The backend for agentic apps.

Define composable agents with built-in memory, tools, sandboxing, orchestration — get a unified API across any model. Spawn one, or thousands.

Get started for free Read the docs

polpo skills loaded

Build a research agent that scrapes papers, summarizes findings, and remembers context across sessions.

1 agent deployed

4 tools registered

Memory + Vault ready

Sandbox runtime live

Application

· designed to work with

· the full stack for ai agents

Everything to run agents in production.

Unified Inference API

Single endpoint with streaming, tool use, and structured output. Works with every LLM via the AI Gateway.

Memory, Skills, Context

Persistent memory, 60+ built-in tools, and extensible skills — agents retain context across sessions.

Sandboxed Execution

Every file write, command, or code run is auto-wrapped in an isolated sandbox with your shared filesystem mounted.

Orchestration Engine

Automatic task dispatch with dependency resolution — tasks execute in the right order and results collect themselves.

Long-running & Scheduled

Agents run for minutes or hours, triggered on a recurring schedule — fire and forget.

LLM-as-a-Judge

G-Eval scoring with custom rubrics — coherence, consistency, fluency, and relevance rated 1-5 automatically.

· benchmark

The cost of shipping an AI agent.

From zero to production — rolling your own vs wiring it up on Polpo.

build from zeroweeks to months

LLM plumbing, streaming, tool-use, memory, sandbox, API, and more…

with polpo30 min

One CLI. Define agents, hit deploy. Infra handled.

System overview

How Polpo fits in your stack.

Your app calls Polpo through an OpenAI-compatible API. Polpo runs your agents in isolated sandboxes, gives them memory and tools, and routes every model call through a unified LLM gateway.

Your App

Web

Mobile

Backend

POST

/v1/chat/completions

agents3 running

social-monitorsandboxtask

reply-draftersandboxtask

escalatorchat

eventslive

tool.web_search · ok

memory.write · +512 tokens

sandbox.spawn · ephemeral

AI Gatewaylive

routing

anthropic

claude-sonnet-4-5

180tok/s

throughput

42ms

latency

99.9%

uptime

available

anthropic

openai

google

xai

mistral

groq

deepseek

cohere

works with any framework

How it works

Three steps to your first agent

Define your agents

Define each agent's harness in a single JSON file — role, model, tools, skills, teams. Version-controlled, portable.

json

[{
  "name": "coder",
  "role": "Senior Engineer",
  "model": "anthropic/claude-sonnet-4-5",
  "systemPrompt": "Write clean, tested TypeScript...",
  "allowedTools": ["bash", "read", "write", "edit"],
  "skills": ["frontend-design", "testing"],
  "reasoning": "medium",
  "reportsTo": "tech-lead",
  "maxConcurrency": 3,
  "identity": {
    "displayName": "Marco Rossi",
    "title": "Full-Stack Developer",
    "tone": "Technical but concise"
  }
}]

Deploy with CLI

One CLI command. Polpo provisions isolated infrastructure and gives you a live endpoint.

shell

$ polpo deploy --dir ./my-project
Deploy from ./my-project/.polpo
Core:
  Teams ........... yes
  Agents .......... yes
  Memory .......... yes
  Skills .......... yes
  Vault ........... yes
  Sessions ........ yes
Deployed: 3 team(s), 8 agent(s), 2 memory file(s),
          4 skill(s), 12 vault entry(ies)

Call from anywhere

One unified API across any model. Your mobile app, web app, CLI, backend — anything that speaks HTTP talks to your agents.

typescript

const res = await fetch("https://api.polpo.sh/v1/chat/completions", {
  method: "POST",
  headers: {
    "Authorization": "Bearer sk_live_...",
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    agent: "coder",
    messages: [{ role: "user", content: "Refactor the auth module" }],
    stream: true,
  }),
});
// SSE stream — same format as OpenAI
for await (const chunk of res.body) {
  // { choices: [{ delta: { content: "..." } }] }
}

~/dev/acme-app — zsh

Use cases

What teams build with Polpo

App builders

Build AI-powered apps with agents as the backend. Ship features, not infrastructure.

like Lovable, Bolt, Replit Agent

AI employees

Agents with identity, memory, and skills that work as autonomous team members. 24/7, no burnout.

Zero-human companies, digital workforce

Voice agents

Plug voice agents into your call center or assistant. Real tools, real actions, real-time.

Support bots, sales assistants

Agent control panels

Build dashboards to manage your agents. Monitor runs, inspect logs, control execution — all via API.

Internal tools, admin panels

Autonomous workflows

Multi-agent missions that run on schedule. Triage tickets, generate reports, monitor systems.

DevOps, support, data pipelines

Portable agents

Spin up disposable agents for batch tasks, benchmarks, or as extensions of your own coding agent. CLI or API, no infra to babysit.

Batch · Benchmarks · Coding-agent extensions

· faqs

Questions, answered.

The short version. For deeper answers, the docs are exhaustive.

read the docs

Polpo is the backend for agentic apps. It runs your agents in isolated sandboxes, gives them memory, tools, skills, and vault, and exposes an OpenAI-compatible API so any app or language can call them over HTTP.

Stop managing infrastructure.
Start shipping agents.

Free tier included. No credit card.
Deploy your first agent in under a minute.

Build your agents