44 Guides Step-by-step Updated May 2026

Guides

Step-by-step playbooks for everything you'd want to do with Sandbox Platform — SSH into your sandbox, give Claude Code its own machine, run 50 browsers in parallel, deploy AI employees that work overnight. Each guide is self-contained, copy-paste ready, and ships with the actual cost numbers.

Product

The feature pages — what the product does, shown with real screenshots and video from the live app. Two layers: the computer, and what you put on it. See the product overview →

Computers for AI Agents
A real computer for your agents — a cloud browser, terminal, or desktop. Instant start, full isolation, per-minute metering with hard cost limits. The compute core everything runs on.
Your Box, Run Anything
Your box, your stack. Install anything from the App Store, self-host your own tools, bring your own Docker, and dev in over a real terminal. Own it all.
Hire AI Employees
Hire a claw from the App Store; it works on a computer you own, remembers its work, and answers you in your channel — and never logs off.
Scheduled Coding Agents
Put your coding agents on a schedule with safe defaults on — PR-not-push, dry-run first, cost caps. Wake up to a reviewable artifact.
Teams That Coordinate
A manager plus workers that share a filesystem and a comms line on your box — one source of truth, not a group chat. Compose the roster once.
Product Overview
The whole product on one page: the computer (browsers, terminals, desktops, computers, blueprints) and what you put on it (apps, employees, teams).
Start Here
SSH Into Your Sandbox
Generate a key, download the .pem, run ssh -i my.pem ec2-user@<ip>. The same flow you already know from AWS, applied to your sandbox — plus the security model, the cert path for power users, and what to do when SSH won't connect.
Read the guide →
 Most Popular
Give Claude Code a Sandbox
Claude Code has 18.9M monthly users but runs on your laptop. Give it a dedicated cloud machine — an isolated Linux environment where the agent lives, works, and persists. SSH in, run claude, walk away.
Read the guide →
Getting Started

Start here if you're new. The mental model and your first 10 minutes.

Setting Up Your Claw
The two credentials a personal AI assistant needs — a Telegram bot token from BotFather and a model API key from Anthropic or OpenAI. Both are yours; here is exactly how to get each and where to paste it.
AI Coding Agents Need Computers: The 2026 Landscape
Claude Code, Codex, Devin, Cursor, and dozens of open-source agents are writing code autonomously. But most run on your laptop. Here's why that's a problem.
Your First AI Employee: From Laptop to Cloud in 10 Minutes
Give the agent its own machine and watch what changes. 10 minutes from "I should try this" to "my agent is running on a cloud computer, working while I make coffee."
What Changes When Your Agent Has a Dedicated Machine
Overnight runs. Parallel execution. Disposable environments. Persistent workspaces. The mental model shift from "AI tool I use" to "AI employee I manage."
Put an AI Coding Agent on a Cron
Hand a coding agent a job and a schedule, and it shows up, does the work on your own box, and leaves you a reviewable result. Managed task workers, from create to review.
The Employee Who Never Logs Off
Always-on AI agents on your own server: a name, a channel, memory that persists, credentials that never leave your box. How a persistent employee differs from a stateless bot.
Give Your AI Agents a Shared Brain
Coordination on your own box: a shared filesystem, an event log, mailboxes and leases, so a team of agents works together without a peer-to-peer mess.
Build a Team of AI Agents That Actually Coordinate
One agent is a worker; a team is a company. How a shared filesystem, a real-time comms line, and safe locks let a manager and workers build together, and how the claws team builder composes the roster.
Building Blocks, Not Bespoke Scripts
How anything gets set up on your box: declarative building-block steps that install their own dependencies, on any machine, without assuming what is already there.
Claude Code Today, Codex Tomorrow
Why your AI worker should not be locked to one brain: harness-as-config, model-key sovereignty, and the roadmap for swapping the agent behind the job.
Framework Integrations

Wire sandboxes into the agent framework you already use.

OpenAI Agents SDK with Sandbox
The OpenAI Agents SDK has 10 million monthly downloads but no built-in code execution. Wire sandbox tools into the SDK so your agent can execute code, browse the web, and manage files.
LangChain Agent with Browser and Terminal
LangChain has 47 million PyPI downloads but no built-in compute. Create LangChain tools for sandbox operations and wire them into a ReAct agent.
CrewAI Crew with Dedicated Machines
Each crew member gets their own sandbox — the researcher gets a browser, the coder gets a terminal, the analyst gets a desktop with LibreOffice.
MCP Server for Any Agent
Build an MCP server that exposes sandbox operations to any MCP-compatible client — Claude Code, Cursor, Gemini CLI, Codex CLI. Write once, every agent can use it.
Browser Automation

Headless Chrome over CDP, fan-out at 50 browsers, vision-model agents.

Playwright Meets Cloud Browser
Launch Chrome containers in the cloud, connect Playwright via CDP, and build browser agents that navigate, extract, fill forms, and take screenshots at scale.
50 Browsers in Parallel
Launch 50 Chrome instances simultaneously. Each navigates a different URL, extracts structured data, and reports back. Fan-out/fan-in with asyncio. Total cost: $0.20.
AI Agent Fills Out Web Forms
Build an AI agent that logs into vendor portals, navigates multi-step forms, fills fields, uploads documents, and submits. Handles sessions, MFA, CAPTCHAs, and error recovery.
Vision Model Browser Agent
Build an agent that takes screenshots of a cloud browser, sends them to Claude or GPT-4o, gets back click coordinates, and executes actions. Works on any website.
AI Employee Use Cases

End-to-end agents that do real work. Cost numbers included.

The AI Research Analyst
Deploy an AI agent that opens 50 browser containers in parallel, reads every page, extracts structured data, and compiles a finished report by morning. Replaces 2 days of analyst work. Cost: $2.50.
The AI Accounts Payable Clerk
An agent fleet logs into 20 vendor portals simultaneously, downloads invoices, extracts line items, and posts to your ERP. Replaces 8 hours/week. Cost: $0.50/run.
The AI QA Engineer
Companies ship 3-5 times a day but test manually once a week. An AI agent tests every user flow after every deploy — no selectors, no maintenance. Cost: $0.15/run.
The AI Competitive Intelligence Analyst
An AI agent monitors 30 competitors daily for $1.50/month — with visual change detection that catches what text-based tools miss. The briefing arrives before standup.
The AI Code Reviewer
For every pull request, an AI agent spins up a fresh sandbox, runs the test suite, visually verifies the UI, and comments with results. Cost: $0.05/PR.
The AI Desktop Worker
$4.2 trillion of enterprise work happens in GUI applications with no API. SAP, Oracle, government portals. Your AI agent can now operate them — seeing the screen and clicking like a human.
The AI Data Processing Pipeline
Upload a CSV, describe the analysis, download the report. AI agent handles data tasks that fall between "too simple for a data team" and "too complex for a spreadsheet." Cost: $0.10-2.00/run.
Managing AI Employees

Hire, onboard, supervise, and review your AI workforce. The operational layer most teams skip.

Onboarding an AI Employee
The day-1 checklist. CLAUDE.md as a job description, Slack integration, trust ladder, 30-minute setup. Everything a new AI employee needs before their first task.
Supervising AI Workers
The daily digest contract, the approval gate, the escalation policy. How to manage 10 AI employees without becoming their full-time babysitter. Includes the morning review workflow.
When AI Agents Fail: Error Handling and Recovery
Every AI agent fails eventually. The question is whether it fails gracefully. Retry logic, circuit breakers, checkpointing, partial-result recovery, and the human escalation path.
AI Agent Performance Reviews
Monthly output audits, CLAUDE.md iteration, trust tier advancement from probation to veteran. The management cadence that keeps AI employees improving over time.
AI Employees: Department by Department
Finance, Legal, Engineering, Customer Support, Marketing, Sales, HR. Where AI employees outperform human admins in 2026, where they don't, and the rollout sequence that works.
Industry Deployments

Vertical-specific AI employees with CLAUDE.md templates, compliance notes, and real ROI tables.

The AI Paralegal
Case law research, discovery review, contract drafting, deadline tracking. Covers citation verification, bar ethics compliance, the 42 states with guidance, and a full Harvey vs Spellbook comparison. 10-attorney firm ROI: $34K/month recaptured.
The AI Listing Coordinator
CMA in 8 minutes, MLS entry, portal cross-posting, open-house follow-up. Fair-housing check built into every listing draft. 10-agent brokerage: 108 hr/month saved at 54× leverage. $150/month.
The AI Practice Manager
Insurance eligibility verification, prior-auth submissions, intake-to-EHR data entry, denial management. HIPAA-aware with BAA available. 4-provider practice: 258 admin hours/month recaptured. $250/month.
The AI Account Manager
CRM hygiene, meeting prep briefs, renewal risk tracking, QBR deck drafts, post-meeting follow-up emails. Connects to Salesforce, Gainsight, Gong, Outreach. 15-person team: 449 hr/month recaptured. $200/month.
Architecture & Operations

How to run agents at scale — events, persistence, parallelism, multi-tenancy, budgets, security.

Running 100 AI Agents in Parallel
Map, Pool, and Pipeline patterns for parallel agent fleets. Full Python with Anthropic SDK, asyncio orchestration, cost math (per-sandbox-second billing), and an end-to-end 200-file PR review example.
Warm Pools: Sub-Second Agent Start
Cold start = 30-45s. Warm pool start = 200-800ms. Snapshot architecture, static/predictive/on-demand pre-warming strategies, break-even cost model, production Dockerfile spec.
Multi-Tenant AI Agent SaaS
Process isolation, filesystem isolation, egress allowlists, per-tenant CLAUDE.md injection, cost attribution, SOC 2 controls table. How Cursor, GitHub Copilot Workspace, and Linear do it.
Persistent Workspaces: AI Agent State
An AI employee that remembers. Persistent filesystem, checkpoint/resume, workspace versioning, tmux sessions that survive reboots. The difference between a tool and a coworker.
Custom Docker Images for AI Agents
Three reference Dockerfiles (coding agent, browser agent, vertical agent), GitHub Actions CI/CD, multi-stage builds, CVE scanning with Trivy, warm-pool pairing. Stop fighting cold-start dependency installs.
Event-Driven Agent Architecture: Deep Dive
The full technical reference: SQS fan-out, SNS topics, webhook ingress, dead-letter queues, exactly-once semantics, and the choreography vs orchestration tradeoff for agent fleets.
AI Agent File I/O: Datasets and Results
Mount S3 buckets, stream large datasets, write structured results back. Efficient I/O patterns for agents that process millions of rows, thousands of files, or multi-GB outputs.
Event-Driven Agents: Work That Starts Itself
AI employees react to emails, calendar reminders, Slack messages, and deadlines — just like human employees. Webhooks, cron, SQS, calendar-aware scheduling.
The Always-On Agent
An always-on agent has an office — files from last week, notes from yesterday, context that makes it better each day. Persistent sandboxes, checkpoint/resume, tmux sessions. Cost: $0.96/day.
Cost Control: Your AI Workforce Budget
50 AI employees running 4 hours/day: $1,320/month total. Per-department budgets, auto-stop, spending caps, and the monthly report template. The guide you forward to your CFO.
Task Design for AI Agents
The #1 reason AI agents fail isn't the model — it's the task description. "Make this better" fails. "Sort column C, add a SUM, export as PDF" succeeds in 30 seconds.
Security for AI Agents in Production
Network isolation, egress control, short-lived credentials, audit logging. What happens when an agent goes rogue? The guide your security team needs.
Strategy

The bigger picture — economics, market landscape, what comes next.

The Economics of AI Employees
An AI employee costs $5-15/day. A human costs $200-800/day fully loaded. But this isn't about replacing headcount — it's about what happens when every employee manages 20 AI workers.
Build vs Buy: AI Agent Infrastructure in 2026
The AI sandbox market has 6 funded startups. Here's the honest landscape: who does what, what it costs to build yourself ($200K+ and 6 months), and when each option is right.
What Comes Next: The Agent-Native Company
Companies with 5 people and 50 AI agents already operate like companies 10x their size. Here's what the org chart looks like in 2028 — and how to be there first.

Stop reading. Start building.

Every guide here is something you can copy-paste-run today. Get an account, generate a key, ssh in.

Open Dashboard