Testing your setup in the AI Playground

The AI Playground in CloudRadial AutomationAI is a test surface for your installed extensions and runners. You tell Orion, the AutomationAI assistant, what you want done, and Orion turns your request into a plan: which runner does the work, which company it is for, which extensions it may use, and whether it will change any data. Orion asks about anything it cannot settle on its own, and nothing is sent to a runner until you review the plan and select Start. The runner then works toward the goal using your own AI provider, your installed extensions and the secrets in your runner's Key Vault, and reports back. Each request is kept as a conversation, so you can follow up on a result without restating it, and convert a conversation that works into a workflow. This article explains how to start a conversation, name a runner or company, review and start a plan, follow a run's activity, and follow up. The AI Playground is available to all roles.

The AI Playground is not a setup guide or a general-purpose assistant. It cannot install extensions or add secrets for you, and a goal that needs a tool you have not installed and configured cannot be planned. To create a workflow, playbook, agent or extension, use the Building something? cards described below.

In this article:

• What the AI Playground Is
• Before You Start
• Starting a Conversation
• Naming a Runner or Company
• Reviewing and Starting the Plan
• Following a Run's Activity
• Charts and Tables in Results
• Following Up in a Conversation
• Building Something Instead
• Planning in Microsoft Teams
• Cancelling a Plan or a Run

What the AI Playground Is

The AI Playground works in conversations. Each conversation starts with a request, and every request in it goes through the same steps:

1. You describe what you want done.
2. Orion builds a plan and asks about anything it needs.
3. You review the plan and select Start. Only then is the work sent to the runner.
4. The runner works toward the goal, and the activity and result appear in the conversation.

Planning uses only what AutomationAI already knows about your workspace (your runners and their status, your companies and their nicknames, your installed extensions, and the secrets your runners have reported). It never calls an extension tool, so nothing in any system is read or changed while you plan.

Open it from AI Playground (/ai-playground) in the sidebar. Your conversations are listed beside the open conversation. On a narrow screen, select Conversations to show the list. When a conversation needs you (Orion has a question, or a running step needs input or approval), the AI Playground sidebar entry shows how many conversations are waiting for you. To find, rerun or archive past conversations, see Reviewing AI Playground history.

Before You Start

The AI Playground runs against what you have already set up, so complete these first:

• A registered runner that is reporting healthy. See Deploying a runner from the Azure Marketplace
• Access to your runner's Key Vault. See Accessing and updating your runner's Key Vault
• An AI provider set on the runner. See Choosing the runner AI provider
• The extensions your goal needs, installed with their secrets in the Key Vault. See Browsing and installing extensions

Starting a Conversation

Opening the AI Playground, or selecting New conversation at the top of the list, shows the new conversation screen. Its introduction links to your installed extensions, your runners and your companies, so you can check what is available before you write a request.

1. In What should Orion do?, describe the task, for example "have @hq-runner list current KnowBe4 users for Acme Accounting".
2. Select Plan or press Enter. Press Shift+Enter to add a new line instead. Planning never starts any work.
3. If Orion asks a question, select one of its quick replies, or type in Or type your answer and select Reply. Orion asks one question at a time.
4. Review the plan, then select Start.

Orion asks only about things that would change the plan or the goal, such as which company you mean, which records or time range, or how the result should be shaped. If everything in your request is clear, Orion asks nothing and the plan is ready straight away. Your answers are added to the goal the runner receives. The conversation's title is a short form of your first request, and you can rename it later.

Naming a Runner or Company

You can say where the work runs and which company it is for in the request itself.

Runners

Type @ to see suggestions grouped into Runners and Companies. Each runner shows whether it is online or offline. Use the arrow keys and press Enter or Tab to insert one, or press Escape to close the suggestions. You can also name a runner in plain words:

• "have <runner> …"
• "ask <runner> to …"
• "on <runner>", "use <runner>" or "using <runner>"
• "<runner>: …"

A runner is recognized only when its name exactly matches one of your registered runners, and a runner name used as an ordinary word elsewhere in a sentence is not treated as a mention. The runner phrase is removed from the goal the runner receives. If you name no runner and your workspace has only one, the plan uses it; otherwise the plan asks you to choose one.

Companies

Name a company with @, or write "for <company>", "at <company>" or "<company>'s". Matching ignores capitalization, punctuation and common legal suffixes such as Inc or LLC, and the company name stays in the goal. If a name matches more than one company, or none exactly, Orion asks which company you mean instead of guessing.

Companies can also have nicknames, so "for AA" can mean Acme Accounting. To add one, open Companies, select the company, enter the nickname under Nicknames and select Add nickname. Each nickname can belong to only one company.

All Companies

Phrases such as "all clients", "every company" or "each customer" set the request to all companies. When a chosen extension can use company-specific credentials, the plan reads All companies, one at a time and the runner works through each company in turn. When every chosen extension uses shared credentials, the vendor's data already covers every company, so the plan reads All companies (one pass, shared credentials) and the runner makes a single pass.

Routing Chips

As you type, chips under the composer preview the result, such as To: with the runner and Company: with the company. A chip with a question mark needs attention, for example when a name matches more than one company. Select a chip to choose a different runner or company, or to clear it. A choice made on a chip replaces the mention in your text.

Reviewing and Starting the Plan

The plan shows one row each for Goal, Runner, Company, Tools, Changes and Mode. Each row says where its value came from, for example "from your prompt", "chosen by Auto", "chosen by you", "your answer" or "from the last step". You can change any of them directly, without answering a question:

• Goal: select Edit prompt, change the text and select Re-plan. This rebuilds the whole plan.
• Runner and Company: pick a new value from the Change… list. The company list includes No company and All companies.
• Tools: select Choose tools, select the extensions the run may use, and select Use these.

The Changes row says whether the run will change data: "Read-only — nothing in any system will be changed", or "This will change data. Each change will still ask for approval." Each change waits in Action Required until someone approves it, unless someone chooses Approve all & future there, which approves every later change in that run without asking again.

Tools and Auto

Unless you choose the extensions yourself, Auto picks them from your goal, and the Tools row shows them as "chosen by Auto" with the number of tools they add. Only the chosen extensions' tools are sent to the model. When the goal would change data, Auto does not decide for you: it lists its picks as Suggested, and Start stays unavailable until you select Use these tools or choose the tools yourself.

What Blocks Start

Start is unavailable, with the reason and a fix shown on the affected row, when:

• No runner is chosen, or the chosen runner is no longer available
• The company is not settled
• No extension fits the goal
• The chosen extensions expose more tools than one run can use
• A chosen extension isn't set up for the company on that runner
• The goal changes data and you have not chosen its tools yourself

Some notes do not block Start. If the runner is offline, the plan says the run starts when the runner comes back online. If the runner has not yet reported a company's secret for a chosen extension, the plan says it is not verified on that runner. Questions Orion marks as optional can be skipped: selecting Start with one still open records it as skipped.

Dry Run and Live

The Mode row shows whether the run is Live or a Dry run. A new plan is Live. Select Switch to Dry run to have changes recorded but not made, or Switch to Live to switch back. A follow-up keeps the mode of the step before it.

Starting

When nothing blocks the plan, it says it is ready to run and that nothing has been sent to the runner yet, and Start is available. Start is the only action that sends work to a runner, and it runs exactly the plan you saw. Start re-checks the plan first. If anything changed since you last saw it (for example, the runner was removed, an extension was uninstalled, or the plan was edited elsewhere), nothing runs. Instead, the plan shows "This plan changed — review it before starting.", marks the changed rows, and the button reads Review and start.

A plan expires after 30 minutes without activity, and the time it expires is shown beside Start. To pick up an expired plan, select Plan again.

Planning and Orion's suggestions use a small amount of AI, which counts toward your workspace's Usage Credits.

Following a Run's Activity

After you select Start, the step appears in the conversation as a block with its goal, its status, and chips for the runner, company, tools ("Auto:" when Auto chose them) and mode. Select Plan on the block to see the confirmed plan with Orion's questions and your answers.

• Until the runner picks the work up, the step reads "Queued — waiting for the runner."
• The Activity list shows reasoning steps as Thinking, tool calls as Tool with the tool name, and previewed changes in a dry run as Dry run.
• When the step finishes, its result and finish time appear in the block.

If the runner needs information or an approval before it can continue, the conversation shows Waiting for you and the item appears in Action Required. See Handling Action Required items in the Inbox. Approvals, denials and answers are recorded in the conversation, wherever they were made. Every step is saved, so a dropped connection never loses the result.

Charts and Tables in Results

When a result is easier to read as a picture, such as ticket trends over the last year or users by company, Orion can answer with a chart. You can also ask for one in your request, for example "show a bar chart of users by company." Charts appear between the text of the result:

• Line charts show change over time
• Column and Bar charts compare values
• Pie charts show the parts of one whole

Each chart has its title as a caption. Select Show data under a chart to see its numbers in a table, and Hide data to close it again. An answer can hold up to three charts. If Orion writes a chart that can't be drawn, the result shows "This chart couldn't be drawn:" with the reason and the chart's raw data, so nothing is hidden. Tables written in a result now display as tables too.

Charts appear only in conversation answers, in the AI Playground and in Orion in Microsoft Teams. Workflow runs, playbooks and Pulsars don't produce them.

Following Up in a Conversation

Use the Next step box at the bottom of a conversation to continue it, for example "now include archived users too". A follow-up starts from the conversation's latest confirmed plan, so it keeps the same runner, company and tools, and the runner receives a summary of the earlier steps and their results. If your follow-up names nothing new, the plan is ready to start straight away. If it names a different company or runner, Orion plans again on what changed.

A conversation runs one step at a time. You can plan the next step while the current one is running, but its Start waits until the current step finishes. A step waiting for input or approval still counts as running.

When a step succeeds, Orion may suggest up to three next requests as buttons under the latest step. Selecting one plans it as the next step; you still review it and select Start. A failed step gets no suggestions.

Only the person who started a conversation can continue it. Other members of your workspace can open it read-only.

Building Something Instead

Conversations run tasks; they do not create workflows, playbooks, agents or extensions. The new conversation screen has a Building something? section with four cards: Create a workflow, Create a playbook, Create an agent and Create an extension. Each card opens that builder with AI Generate already open, so you can describe what you want and review what it builds.

Planning in Microsoft Teams

The same planning works when you message Orion in Microsoft Teams. Orion's questions appear as card buttons, the plan arrives as a card with Start, and nothing runs until you press it. Results with charts show them as Teams charts on the run card. Conversations started in Teams appear in your AI Playground list marked Teams, and you can continue them on the web. See Using Orion in Microsoft Teams.

Cancelling a Plan or a Run

To drop a plan you no longer need, select Cancel beside Start. Nothing was sent to a runner, so nothing needs undoing. Once a step has started, the conversation has no control to stop it. If the step pauses for input or approval, you can cancel it with Cancel run in Action Required. Cancelled runs remain in your history with their audit.

If you are still having trouble, we're here to help! Submit a ticket for assistance, and don't forget to check our status page to ensure there are no outages in your area.