Key facts
| What it is | Browser-based sovereign AI agent playground |
| Availability | Beta |
| Input | A plain-language goal |
| Visibility | Live step tracker showing plan and progress |
| Oversight | Pause for approval before risky actions |
| Infrastructure | Runs on sovereign infrastructure you control |
| Entry | Free plan with plugsky-micro and plugsky-lite, no card |
TL;DR
- Type a goal and watch a sovereign agent plan and execute it.
- Sub-agents and tool calls are visible in a live step tracker.
- The run pauses for approval before risky actions.
- Your model, your region, your audit log — no third-party inference.
- Beta software: evaluate before production use.
How it works, step by step
- Open the Plugsky Playground Beta in your browser; no install is required.
- Enter one bounded goal with a clear artifact as the outcome.
- Watch the agent decompose the goal and choose tools in the step tracker.
- Approve or reject the action when the run pauses at a risky step.
- Inspect the run history and audit entry when the task completes.
- Repeat with a harder goal, then decide what needs a human gate in production.
Try it yourself
What you can try today
The playground is a working environment, not a slide deck. Enter a goal and the agent proposes a plan, breaks it into steps, and starts executing with the tools it has been given. You can watch it spin up sub-agents for parallel work, run code, and assemble a result. Every step is visible, which is the point: before you delegate real work, you should see how the agent behaves.
The live step tracker
Most agent failures are invisible in chat output. The step tracker exposes the run as it happens: which step is active, which tool was called, what came back, and where the plan changed. That view is what makes evaluation possible. You can spot a bad assumption in step two instead of discovering it in a finished artifact, and you can compare two runs of the same goal to see where behavior diverges.
The trade-confirmation gate in action
When the agent reaches an action classified as risky — moving money, deleting data, sending externally — the run pauses and asks. You approve, reject or adjust before anything irreversible happens. This is the same gate pattern you would wire into production, so testing it in the playground tells you whether your thresholds and approval flow are practical for real operators.
Your model, your region, your audit log
The playground runs on sovereign infrastructure rather than a third-party consumer service. That means you can evaluate agent behavior without shipping sensitive prompts to a black box, and the run history stays under your control. For regulated teams, this is the difference between a demo that security blocks and a pilot they can approve. Check the docs for current deployment and logging options.
Join the beta
The Playground is live in beta, so expect rough edges and changing behavior. Bring a real but low-risk task, note where the agent needs better tools or tighter permissions, and use those notes to scope a production pilot. Create a free account to start; the free plan includes plugsky-micro and plugsky-lite with no card, and paid tiers can be evaluated with a 14-day full-access trial.
Honest comparison
| Dimension | Plugsky Playground Beta | Typical agent demo | Production agent platform |
|---|---|---|---|
| Access | Browser, no install | Guided demo | Account and setup |
| Plan visibility | Live step tracker | Highlights only | Logs and dashboards |
| Tool use | Real tool calls in a scoped run | Simulated | Full integration |
| Approval gates | Interactive pause and approve | Rarely shown | Configurable policy |
| Data path | Sovereign infrastructure | Vendor cloud | Your deployment |
| Maturity | Beta | Demo | Production |
Frequently asked questions
Is the Plugsky Playground production-ready?
No. The playground is in beta and meant for evaluation. Use what you learn there to scope a pilot, then move to a supported deployment for production traffic.
Do I need to install anything?
No. The playground runs in your browser. If you later want to call the agent programmatically, use the OpenAI-compatible API and your existing SDK.
Which models power the playground?
It runs on Plugsky models behind our OpenAI-compatible API, with 30+ models available across tiers. See the live model catalogue for current names and capabilities.
Can I see why the agent made a decision?
Yes. The step tracker shows the plan, tool calls and results, and the run produces an audit entry you can inspect afterwards.
Does the playground send my data to third parties?
Runs execute on sovereign infrastructure, and data stays within your chosen jurisdiction. Review the docs and legal terms for the current data-handling details.
What does it cost to try?
You can start on the free plan with plugsky-micro and plugsky-lite, no card required. A 14-day full-access trial covers the paid tiers; see the live pricing page.
How is beta feedback handled?
Treat beta behavior as subject to change, and report issues through the usual support channels so they can be triaged before general availability.
Plugsky (2026). “Try a Sovereign AI Agent — Playground Beta”. Plugsky. Available at: https://plugsky.com/news/ai-agent-playground-beta (last updated 2026-09-25).