Give AI agents a computer, a plan
and the tools to finish work.

Computer Agents lets you run AI agents like a real team. Each agent gets a persistent cloud computer, project tasks, memory, files, skills, subagents, and deployment tools, so it can research, build, operate, and improve products and workflows from start to finish.

Trusted by leading teams around the world

Nvidia logo
PioneerAI
FireChatbot logoFireChatbot

Work does not disappear
when the chat ends.

Computer Agents keeps project plans, tasks, files, agent runs, cloud computers, and triggers connected in one persistent workspace, so agents can continue where they left off, ship real deliverables, and respond to new events over time.

A real computer for every agent.
Browser, terminal, files, GUI.

Each agent gets its own persistent cloud computer, so it can browse the web, run code, edit files, install packages, and create real outputs instead of stopping at a text response.

ACP workspace preview

Trigger from anywhere.
Email, schedules, webhooks.

ACP can be activated by outside systems and continue work inside the same persistent context instead of starting from zero.

Email-triggered threadslive

Start work from inbound email and reply with results after the run completes.

Recurring schedulescron

Run recurring project workflows on calendar-driven schedules without separate queueing logic.

Webhook automationsevent-driven

Turn external product events into stateful ACP execution inside threads and projects.

This isn't a chatbot.
It's your AI coworker.

Chatbots forget everything when you close the tab. Your coworker gets its own computer in the cloud with persistent memory, scheduled tasks, and access from any device.

aiOS Central

Same agents. Same work. Everywhere.

Start on your phone, continue on Mac, manage from the web, or build through the API. Your agents, files, tasks, memory, and cloud computers stay in sync across every surface.

Computer Agents across devices

Built with Security.

Isolated execution, explicit data controls, and transparent documentation for teams that need to validate risk before adoption.

Isolated execution

Each agent runs in its own isolated cloud container with strict resource and access boundaries.

Encrypted by default

Data is protected in transit and at rest, with layered authentication and access controls.

Data ownership and controls

You control your threads, files, and environments with deletion workflows and clear retention policies.

Used by Professionals.

Business owners, researchers, and marketers deploy agents that run autonomously, producing real results instead of chat replies.

Architecture
Empowered by AI.

A London-based architect uses aiOS agents to generate professional floor plans in minutes. From concept to client presentation.

Professional Floor Plans
Minutes, Not Hours
Client-Ready Presentations

Customer Support
powered by aiOS.

FireChatbot uses aiOS SDK and Mail integration to build custom Customer Support Chatbots and Email Assistants.

Handles real customer Emails
35% reduced Support Workload
80% Automation Rate

Loved by Developers.

npm install computer-agentspip install computer-agents

Build with official TypeScript and Python SDKs for persistent threads, computers, projects, published resources, webhooks, schedules, state management, and agent email triggers.

Custom Agents

Define agents with custom system prompts, model preferences, and tool configurations in seconds. Each agent gets its own isolated computer environment ready to execute tasks autonomously.

Create your first agent

Persistent Workspaces

Every agent gets a dedicated workspace where files persist across sessions. Durable storage that survives restarts and scales automatically.

Explore workspaces

Code Execution

Run code in fully isolated containers with automatic dependency management. Agents can write, execute, and iterate without custom sandbox orchestration.

Learn about environments

Skills System

Extend agent capabilities with built-in and custom skills. Combine prompt logic, MCP tools, and reusable execution behaviors inside the same platform.

Browse skills

Multi-turn Threads

Maintain context across unlimited turns with persistent threads. Agents remember prior instructions, generated files, and execution history for continuous workflows.

Understand threads

Agent Orchestration

Coordinate multiple agents in workflows. Run steps sequentially, in parallel, conditionally, or map-reduce with shared project context.

Build multi-agent workflows

Projects

Use mission control, backlog, board, releases, and shared attachments to coordinate agent execution inside one product workspace.

Explore projects

Publish Resources

Ship web apps, APIs, functions, auth services, databases, and agent runtimes from the same ACP project where work happens.

See resource model

Webhook Triggers

Fire agents automatically from product events and external systems with signed webhook endpoints and persistent ACP context.

Set up webhooks

State Management

Fork environments and threads, inspect diffs, and continue from prior execution states instead of restarting from scratch.

Learn state management

Schedules

Run threads, agents, and project workflows on cron schedules without building your own queue or scheduler infrastructure.

Set up schedules

Agent Email Addresses

Give agents their own email address so inbound email can activate threads, carry attachments into the default environment, and reply with run summaries.

See integrations

Configure your Containers

Pick Lite, Standard, Power, or Desktop, install any software you need, configure runtimes, packages, secrets, and MCP tools, and shape the execution environment around the work instead of forcing the work into one fixed runtime.

Read Docs
Computer container size options

Deploy Resources

Publish web apps, APIs, functions, auth, databases, and agent runtimes from the same ACP workspace where threads, files, agents, and project execution already live.

Read Docs
Deploy resources

$20/month for your first
AI employee.

Start free with 150 Compute Tokens. Upgrade when your agents prove their worth.

Compute Tokens in Real Terms: Tasks

Pick a task type and instantly see how many task starts each plan can handle per month.

Small follow-up using Claude Haiku 4.5 on the Lite runtime profile. Estimate basis: ~4 CT per task start on the Lite computer profile.

Free

150 Compute Tokens
~31-46 task starts

Individual

1.5k Compute Tokens
~312-468 task starts

Team

5k Compute Tokens
~1041-1562 task starts

Enterprise

15k Compute Tokens
~3125-4687 task starts