Build agent harnessesyou run in your app

The development platform for AI agents. Harness, sandboxes, memory, channels and schedules, production-ready behind one API.

Paste into your coding agent

Run npx @polpo-ai/cli create and follow the steps to create my first agent.

Docs

· platform

Everything to run production agents

Agents

A fully composable harness: prompt, skills, and tools, with versions you can review and roll back.

⟲ rollback to v11

agentsupport-agentv12

promptskillstools

Model Gateway

All models, one bill. Route and switch models without rewriting.

POST/v1/chat/completions+12

Tool Functions

Define functions your agent can call as a tool.

defineTool({
  name: "refund",
  run: async ({ id }) =>
    refunds.create(id),
})

agent called refund()

Memory

Agents remember facts and context across runs.

memoryuser-425 facts

Sandboxes

Built-in isolation that automatically spins up when needed.

Durable Loops

Run long-running, deterministic workflows that survive restarts, retries, and waits.

taskenrich-leadsstep 5 / 8

retry3 attempts · backoffresumed after restart

Connections

Give agents the outside world over MCP, OAuth, and API keys.

MCP serversregistry · remote

OAuthscoped · per user

API Keyserver-to-server

Schedules

Run agents on a clock with cron or plain language.

billing-syncevery 15m

inbox-triage09:00 daily

nextin 4m

Channels

Let users talk to your agents from the tools they already use.

plusweb chat · REST API

Drives

Give agents persistent files they can create, read, and reuse across runs.

bucketagent-files

report.pdf2.4 MB

export.csv18 KB

Use one or all. It's one platform.

Trusted by teams shipping agents

Open Magi

Superlog

Open Magi

Superlog

Open Magi

Superlog

· how it works

Keep building from the tools you already use

Install the Polpo MCP or call the API, and ship a consumer-facing agent that lives inside your app.

Plugs into your coding agent

polpo.sh/agents

Agents

support-agent

support

EditRun

AgentChannelsIntegrationsSessionsTools & MCPsSkills

PromptModels

System promptv12

You are Aria, a customer-support agent for an e-commerce store. Resolve order tickets end-to-end: look up orders with check_order, draft a clear, friendly reply, and escalate only when a refund exceeds $200 or the customer is upset. Always confirm the order number before acting, and keep replies under four sentences.

· agents as files

Your agent is
a directory.

Everything the agent is — model, prompt, tools, loops, limits — lives in .polpo/: plain files you read, diff and version like the rest of your codebase. Portable and framework-agnostic, so it never locks into a stack.

agents/support/config · instructions

project.jsonproject & gateway settings

loops/durable multi-step workflows

tools/your own code, sandboxed

Use Polpo from any stack

run-agent.tsone HTTP API

import { Polpo } from "@polpo-ai/sdk";

const polpo = new Polpo();
const run = await polpo.agents.run("support", {
  input: "My subscription did not renew."
});
console.log(run.output);