MIOSA Documentation

Build, preview, deploy, and operate AI-generated applications from one API. MIOSA gives your product isolated sandboxes for code work, live previews for iteration, durable deployments for production, desktop computers for browser agents, and white-label controls for customer-owned experiences.

What can I build? → Start building API reference

Last updated: June 9, 2026

Start here

See concrete MIOSA product patterns: AI app builders, artifact factories, browser agents, vertical SaaS copilots, and durable deploys.

Map Polsia-style operators, Cofounder (cofounder.co)-style company OS products, and Nebula-style virtual devices onto MIOSA.

Link and map Lovable, Replit, Bolt, v0, Genspark, Cofounder, Polsia, Nebula, Heuresis, BusinessOS, and open agent-workspace repos to MIOSA implementation patterns.

Understand the core MIOSA primitives: agents, runtimes, devices, sessions, events, artifacts, deployments, and credits.

Create a sandbox, run a command, expose a live preview, and understand the resource model before wiring MIOSA into your product.

Use isolated Linux environments for generated code, dependency installs, file edits, build commands, tests, and preview servers.

Publish generated applications to stable URLs with versions, releases, environment variables, dynamic runtimes, and rollback paths.

Put previews, deployments, browser tokens, and customer attribution behind your own product and domain model.

Build the server-side control plane that dispatches prompts into sandboxes, computers, and local devices.

Give agents a full Linux desktop when they need browsers, screenshots, clicks, uploads, form fills, or human handoff.

Use the REST API directly, ship with SDKs, or automate workflows from the CLI. Always trust URLs returned by the API.

What you can build

Build prompt-to-app products where agents write code inside persistent sandboxes, show previews, stream command work, version progress, and publish.

Coordinate sandbox agents, computer agents, artifact generators, browser QA, deployment agents, and live UI events.

Choose the right pattern for one-shot runs, sandbox prompts, computer prompts, detached sessions, app builders, and fleets.

Add a backend that resolves users, dispatches prompts, streams agent work, stores artifacts, and enforces quota.

Build autonomous company operators, agent departments, virtual devices, miniapps, jobs, and connected-tool workflows.

Use MIOSA BusinessOS as the open reference for a company workspace with OSA, CRM, projects, docs, calendar, context, modules, and terminal work.

Build shared persistent devices with files, browser work, jobs, artifacts, connectors, credits, and approvals.

Give each customer isolated workspaces, branded preview domains, deployment domains, custom domains, browser tokens, and usage attribution.

Turn prompts into published landing pages, intake forms, calculators, content hubs, and campaign sites without exposing MIOSA.

Run apps that need a server process, API routes, background work, environment variables, and persistent release management.

Execute user code in isolated environments, stream stdout and stderr, capture artifacts, and shut work down when finished.

Use Computers for real browser sessions: navigate, click, type, upload files, capture screenshots, and hand control to users.

Bring third-party computer capacity into MIOSA while keeping one API, one product surface, and one user relationship.

Measure sandbox lifecycles, desktop readiness, screenshot/action latency, and task success across agent workflows.

Platform guides

Concepts Core resource model: workspaces, projects, sandboxes, previews, deployments, computers, and attribution.

Authentication API keys, browser-safe tokens, and customer-scoped access patterns.

Deployment domains Preview domains, deployment domains, custom domains, DNS verification, and TLS.

Data and persistence Postgres, storage, Redis, volumes, and generated app state.

Security Secrets, connect accounts, network rules, audit logging, and white-label credentials.

Evaluations OSWorld-style task scoring, desktop readiness, traces, and regression checks for agent workflows.

Benchmarks Current sandbox and desktop timing measurements, with clear labels for sample size and run context.

Provider comparison How MIOSA, native computers, BYOC, and external providers fit different product needs.

Developer surfaces

Surface	Use it when
REST API	You want complete control over sandboxes, files, exec, previews, deployments, computers, domains, tokens, and events.
Agent product model	You need the canonical architecture for agents, devices, runs, events, artifacts, and credits.
Agent workflow patterns	You want to map user prompts to one-shot runs, sandbox work, computer work, detached sessions, or fleets.
Backend orchestration	You are building the server-side layer that dispatches prompts into MIOSA devices and streams work back to your UI.
Agent credit metering	You need to charge credits for model usage, device runtime, tools, storage, and deployments.
CLI	You want to create, inspect, publish, and debug resources from a terminal.
Python SDK	You are building agent loops, backend orchestration, or data workflows in Python.
TypeScript SDK	You are building Node services, web backends, or product integrations in TypeScript.
MCP server	You want an AI agent to use MIOSA tools through the Model Context Protocol.
Agent Development Kit	You want higher-level agent building blocks on top of the raw sandbox and deployment APIs.

For LLMs

MIOSA publishes machine-readable docs indexes so agents can retrieve current public documentation without scraping the site.

LLMs index Compact public docs map for agent retrieval.

Full LLM context Expanded public docs corpus for deeper agent context windows.

Current production contract

Use URLs returned by the API instead of constructing hostnames yourself:

preview_url for sandbox previews.
desktop_url for computer desktops.
public_url for deployments.
custom_domain or deployment-domain URLs when a workspace has branded domains configured.

See the changelog for the latest public platform behavior.