A “computer-use agent” from Simular Inc. in the United States. Rather than an AI that only writes text in a chat window, Sai takes over an entire dedicated desktop in the cloud (or your own PC) and works by clicking, typing, and scrolling through browsers, business applications, and desktop tools exactly as a person would. Its defining strength is that anything you can operate on screen can be automated, even apps with no public API, so it can handle cross-application routine work such as sorting email, submitting job applications, doing research, and organizing data. Processing continues on the cloud desktop even after you close your laptop, so it is designed to work as an always-on “AI coworker.” The technical foundation is Agent S, the open-source GUI agent framework published by Simular.
Key Features
- Operates the GUI directly: Because it works through the screen rather than API integrations, in-house systems and legacy business apps with no API can also become targets for automation
- Always-on cloud desktop: Tasks run on a dedicated cloud desktop and keep going even when your local PC is off. You can check progress and send instructions from iMessage or Telegram
- Works on your own device (BYOD): You can use a local macOS or Windows environment as the agent’s workspace, which helps when the app you want automated exists only on your machine
- Approval-based safety design: High-impact actions such as sending, purchasing, and deleting are designed to require your approval before execution
- Light setup and persistent memory: Connecting apps is mostly a matter of signing in, and logins, settings, and preferences persist across sessions. Repeated tasks settle into a routine over time, becoming faster and cheaper to run
- Execution engine derived from Agent S: Sai comes from the lineage of Agent S, Simular’s Apache-2.0 GUI agent framework. Its latest version, Agent S3, scored 72.60% on the OSWorld desktop-operation benchmark and was announced as the first to surpass human-level performance (about 72%)
Pricing
| Plan | Monthly price | What’s included |
|---|---|---|
| Starter | $50 | $50 in monthly credits, cloud desktop (sleeps when idle), BYOD support, community support |
| Premium Starter | $200 | $100 in monthly credits, always-on desktop, native app automation, priority support |
| Pro | $500 | Unlimited credits, team sharing, full API access, priority support |
| Enterprise | Contact sales | Managed scaling, dedicated account team, SSO / RBAC, SOC 2, SLA |
Pricing is as of August 2026. Discounted early-access pricing may be offered separately, so the amount actually applied can vary depending on when you sign up. Check the official site for current pricing.
Pros & Cons
✅ Pros
- Automates apps and in-house systems that have no API, so you can adopt it without rebuilding existing workflows
- Work continues on the cloud desktop, so you can hand off long-running tasks and walk away
- No API keys or scripting required, making it approachable for non-engineers
- Approval steps on important actions limit the damage from mistaken operations
- The underlying execution framework (Agent S) is open source, so you can inspect how it works
⚠️ Cons
- Starting at $50 per month even on the entry plan, it is expensive for casual individual use
- GUI operation is sensitive to what appears on screen, so app UI updates or layout shifts can break a procedure
- Keeping business account logins on a dedicated desktop requires alignment with your organization’s information-management rules
- There is limited public evidence of how well it handles Japanese-language business apps, so test it against your own target apps first
- Compared with API-based automation, each task tends to take longer to run
Comparison with Similar Services
| Item | Sai | ChatGPT Agent | Claude (Computer Use) | Manus |
|---|---|---|---|---|
| Provider | Simular Inc. | OpenAI | Anthropic | Monica (Butterfly Effect) |
| Scope of operation | Entire cloud or local desktop | Mainly cloud browser and terminal | Screen operation in a developer-provided environment | Cloud browser and tools |
| Apps without an API | Can operate them | Limited support | Depends on the environment | Limited support |
| Always-on | Continues on the cloud desktop | Per session | Depends on the implementation | Per session |
| Main users | Practitioners and teams offloading routine work | General users to business | Developers | General users to business |
Who Is It For
- Staff who want to automate in-house systems and business apps that offer no API
- Business professionals with weekly recurring work such as reporting, scheduling, and lead sourcing
- People who want to clear out screen-by-screen grind work like job applications and research
- Small teams and solo operators who want to hand off long-running tasks and only receive the results
- Engineers who want to evaluate GUI agents down to the open-source implementation (Agent S)
Summary
Sai moves a step beyond “have the AI write text” toward “have the AI operate the computer itself.” Because its screen-operation approach does not assume API integrations, it brings business applications that were previously out of reach into scope for automation. On the other hand, the pricing is set at a level aimed at business use, and the instability inherent to GUI operation remains. A realistic approach is to pick one target app, start small with approvals in place, confirm that it holds up under repeated runs, and then widen the scope.