> ## Documentation Index
> Fetch the complete documentation index at: https://agentrein.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.
# Introduction
> AgentRein is the reliability, rollback, and governance layer for production AI agents.
## Zero-Friction Reliability for Production AI Agents
AI agents interact with real-world infrastructure by executing sequences of API calls — creating GitHub issues, sending Slack notifications, creating Stripe invoices, and updating databases. However, when an agent encounters an error, experiences goal drift, or fails midway through execution, systems are left in a **partial, corrupted state**.
**AgentRein** provides a safety wrapper around your agent's existing tools and SDKs. Without requiring custom connector classes or modifying your existing codebase, AgentRein intercepts tool executions, logs side-effects, applies governance policies, and enables **automatic Last-In-First-Out (LIFO) rollback**.
If an action fails, an approval is rejected, or an error occurs midway, AgentRein automatically reverses every previous production action in strict LIFO order.
Pause execution on sensitive operations (like deleting databases or canceling subscriptions) until approved by a reviewer via the dashboard or API.
Validate agent output against intended goals using LLM-powered outcome verification to catch drift before it impacts downstream systems.
Enforce organization-level rules (`ALLOW`, `REQUIRE_APPROVAL`, `BLOCK`) across tools with wildcard API pattern matching and payload field conditions.
***
## Bring Your Own Tools & Standard SDKs
AgentRein requires **no proprietary connector frameworks** or wrapper SDK replacements. You continue using standard TypeScript or Python libraries — such as `@octokit/rest`, `stripe`, `@slack/web-api`, or `google-api-python-client` — and wrap them using a single method: `agentrein.wrap()`. For supported actions registered under your connector, AgentRein automatically tracks side-effects, applies governance policies, and prepares inverse rollback steps.
```typescript TypeScript theme={null}
import { AgentRein } from 'agentrein';
import { Octokit } from '@octokit/rest';
import { WebClient } from '@slack/web-api';
const agentrein = new AgentRein({ apiKey: process.env.AGENTREIN_API_KEY });
const octokit = new Octokit({ auth: process.env.GITHUB_TOKEN });
const slack = new WebClient(process.env.SLACK_TOKEN);
// 1. Start a tracked session with an explicit intent
const session = await agentrein.newSession({
agentId: 'incident-responder',
intent: 'Create incident tracking issue and notify the engineering channel',
});
// 2. Wrap standard clients with AgentRein
const agentOctokit = agentrein.wrap(octokit, session, { connector: 'github' });
const agentSlack = agentrein.wrap(slack, session, { connector: 'slack' });
try {
// 3. Execute calls naturally — side-effects and rollbacks are tracked automatically
const issue = await agentOctokit.issues.create({
owner: 'acme-corp',
repo: 'infrastructure',
title: 'High Memory Usage Alert',
});
await agentSlack.chat.postMessage({
channel: '#incidents',
text: `Created GitHub issue: ${issue.data.html_url}`,
});
// 4. Mark session complete when successful
await agentrein.completeSession(session);
} catch (err) {
// Automatic LIFO rollback has already executed for all logged actions in this session!
console.error('Session failed and rolled back:', err);
}
```
```python Python theme={null}
import os
from agentrein import AgentRein
from github import Github
from slack_sdk import WebClient
agentrein = AgentRein(api_key=os.environ["AGENTREIN_API_KEY"])
gh = Github(os.environ["GITHUB_TOKEN"])
slack = WebClient(token=os.environ["SLACK_TOKEN"])
# 1. Start a tracked session with an explicit intent
session = agentrein.new_session(
agent_id="incident-responder",
intent="Create incident tracking issue and notify the engineering channel",
)
# 2. Wrap standard clients with AgentRein
repo = gh.get_repo("acme-corp/infrastructure")
agent_repo = agentrein.wrap(repo, session, connector="github")
agent_slack = agentrein.wrap(slack, session, connector="slack")
try:
# 3. Execute calls naturally — side-effects and rollbacks are tracked automatically
issue = agent_repo.create_issue(title="High Memory Usage Alert")
agent_slack.chat_postMessage(
channel="#incidents",
text=f"Created GitHub issue: {issue.html_url}",
)
# 4. Mark session complete when successful
agentrein.complete_session(session)
except Exception as err:
# Automatic LIFO rollback has already executed for all logged actions in this session!
print(f"Session failed and rolled back: {err}")
```
**Preserves Native Types & Syntax**: The `wrap()` method uses a dynamic proxy under the hood. IDE autocompletion, type definitions, and original SDK methods remain 100% intact, with full snapshot tracking and rollback protection enabled for all supported connector actions.
***
## Testing Safely with Sandbox Mode
Test your agents without touching production APIs. By using a **Sandbox API Key** (`ar_test_...`), calls wrapped by AgentRein simulate execution against connectors — ensuring zero side-effects on real GitHub repositories, Stripe balances, or Slack channels while fully verifying your session logic and approval workflows.
```bash theme={null}
# Switch seamlessly between environments:
AGENTREIN_API_KEY=ar_test_51Nx... # Development Sandbox
AGENTREIN_API_KEY=ar_live_98Zb... # Production Environment
```
***
## Explore the Documentation
Get AgentRein setup in your TypeScript or Python agent in under 5 minutes.
Learn how the Transparent Proxy, Session State Machine, and LIFO Engine work.
Explore full parameter specifications for `newSession()`, `wrap()`, policies, and error handling.
This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.