---
title: "AgentTwin | Digital twins for testing AI agents"
description: "AgentTwin builds digital twins for testing AI agents. Teams prepare repeatable test worlds with synthetic service data, supported failure conditions and explicit outcome checks."
canonical: "https://agenttwin.si/"
---

# AgentTwin | Digital twins for testing AI agents

AgentTwin builds digital twins for testing AI agents. Teams prepare repeatable test worlds with synthetic service data, supported failure conditions and explicit outcome checks.

AgentTwin builds repeatable test worlds for workflows, service interactions and outcome checks. Find candidate code paths in your repository, prepare a test, introduce supported failures and inspect the recorded result.

## Current execution coverage

The guided execution path currently supports prepared Node.js workflows with synthetic data and supported service operations. Discovery alone does not make a workflow runnable. Local execution can access the host filesystem and network. E2B cloud execution is optional. The browser demo uses scripted outcomes. See [current coverage](https://agenttwin.si/docs/status/) for implementation details.

## When to use AgentTwin agent testing

Check final-state assertions or retry safety after a write times out. Replay a captured result or save a reviewed scenario as a regression test. Review operation contracts first. The test world covers declared behavior, not every production condition.

## Connect your project

```sh
npx agenttwin
```

Run in your project folder. Connect your account, then keep the terminal open while requesting preparation from the dashboard. Review the proposed test locally before running it. Follow [project setup](https://agenttwin.si/docs/quickstart.md) for requirements and each next step. Guided tests use local Docker; direct `run` defaults to host execution.

## Optional sample test

```sh
npx agenttwin example --out ./agenttwin-example
cd agenttwin-example
npx agenttwin run --seed 00000000000000000000000000000001 --fault none
npx agenttwin run --seed 00000000000000000000000000000001 --fault timeout-after-write
```

The clean run expects one ticket and one notification. The faulted run exposes duplicate tickets and should fail. See the separate [sample walkthrough](https://agenttwin.si/docs/example.md) for verification, reset and replay. Review files before execution.

## Integration entry points

- [Public documentation MCP](https://agenttwin.si/mcp/docs): search and read guides over Streamable HTTP; no account access or execution.
- [Official public agent tools](https://github.com/AgentTwin/agenttwin-agent-tools): skills, AGENTS.md, Cursor rules and portable plugin configuration.
- [Agent tools on npm](https://www.npmjs.com/package/agenttwin-agent-tools): reviewed source bundle metadata; inspect its skill files before installing.
- [Local synthetic sandbox](https://agenttwin.si/sandbox.md): no account or provider credentials.
- [API versioning and deprecation](https://agenttwin.si/api-policy.md).
- [Agent skills](https://agenttwin.si/.well-known/agent-skills/index.json): prepare real workflows and investigate evidence.

- [Developer guide](https://agenttwin.si/developers.md): npm CLI, local MCP stdio and the documented REST subset.
- [Authentication](https://agenttwin.si/auth.md): account login and terminal pairing; authenticated customer resources.
- [OpenAPI](https://agenttwin.si/openapi.json): scoped OAuth, project metadata, test drafts and reviewed cloud test jobs.
- [API catalog](https://agenttwin.si/.well-known/api-catalog): API descriptions and health endpoint.
- [Operation contracts](https://agenttwin.si/docs/provider-operations.md): supported subsets and limitations.
- [Pricing](https://agenttwin.si/pricing.md): One enterprise offering. Contact us to discuss requirements and pricing.
- [CLI reference](https://agenttwin.si/docs/cli.md) and [npm package](https://www.npmjs.com/package/agenttwin). Implementation source is currently private.
- [About](https://agenttwin.si/about.md), [Contact](https://agenttwin.si/contact.md), [Privacy](https://agenttwin.si/privacy.md).
