Agent Studio - FAQ

Get answers to frequently asked questions about Agent Studio, including genies, knowledge bases, skills, and security features.

Overview

What is Agent Studio?

Agent Studio is where you build and configure AI agents (genies). These genies perform actions, call workflows, understand context, and execute pre-defined skills to achieve your defined goals.

What are genies?

Genies are AI-powered agents built in Agent Studio that pursue goals you define, adapt to context, and act across apps and data systems.

Genies operate through the following key components:

• AI model and job description combine to form the brain and instructions of the genie. These components use LLMs to interpret requests, analyze context, make decisions, generate responses, and define your genie's behavior, persona, and constraints. Genies use Anthropic Claude by default. You can switch your LLM to OpenAI GPT or your own LLM connection.
• Chat interface provides the user interface where users can converse with the genie. You can use Slack, Microsoft Teams, or Workato GO as your chat interface.
• Knowledge base stores company-specific information, conversation history, and metadata.
• Skills enable interaction with various applications and systems.

Setup and access

Who can access Agent Studio and what are the prerequisites?

Genies are an Agentic feature available to customers on specific pricing plans. Genies are available in the US, EU, AU, SG, and JP data centers. Genie models are hosted in the US, EU, and APAC regions and respect data residency requirements where possible.

You must ensure that genies are enabled in your account before using this feature. Contact your Customer Success representative to enable genies if you don't see the genies option in your workspace or require additional information. Refer to your pricing plan and contract to learn more.

Can I enable Agent Studio for child workspaces independently?

No. Agent Studio becomes available to all child workspaces when it's enabled on an AHQ parent workspace. You can't enable Agent Studio for specific child workspaces without enabling the parent first.

You can use collaborator permissions to control which users can access Agent Studio in each workspace.

Where do I access and manage genies?

You can view and configure genies by going to AI Hub and clicking the Genies tab. Refer to Getting started with genies for more information.

What LLM models does Agent Studio support?

Agent Studio supports three LLM options for genies:

• Anthropic Claude (default)
• OpenAI GPT
• Use your own LLM connection (BYOLLM)

Note that the AI model can only be changed after you stop your genie.

What chat interfaces can be used with genies?

You can deploy genies to the following chat interfaces:

• Slack
• Microsoft Teams
• Workato GO

Note that you can't change the chat interface after a genie is created.

What file and image types can genies handle?

Genies support file and image uploads up to 25MB.

Common document file types are supported, including .pdf,.doc and .csv. Refer to Files for a complete list of supported file types.

Common image formats are supported, including .jpg and .png. Video files aren't supported. Refer to Images for a complete list of supported image formats.

How do I create a genie?

You can create a genie in two ways:

• Create a genie from a template
• Create a genie manually

What are the required components to run a genie?

Configure the following components to get your genie working:

• Job description: Define your genie's role and goals.
• Chat interface: Choose where users interact with the genie. Options are Slack, Microsoft Teams, or Workato GO.
• AI model: Select the LLM that powers the genie. Options are Anthropic Claude, OpenAI GPT, or your own LLM connection.
• Knowledge base: Configure knowledge sources for your genie.
• Skills: Enable your genie to take actions.

Refer to Create a genie manually for a step-by-step walkthrough of creating all components together.

What metrics can I track for my genies?

The genie Overview page tracks:

• Total Conversations: Distinct conversations initiated each day
• Total End-User Messages: Message volume to gauge engagement
• Unique Users: Distinct users for measuring adoption
• Response Time: Average, median, and 90th percentile response times
• Conversation Volume: Conversations over time
• Skills Usage Heat Map: Skill execution frequency and success rates

You can filter metrics by time range.

Note: Genies using Workato GO can also track custom business KPIs. Refer to the Action Board section for more information.

Knowledge bases

What is a knowledge base?

A knowledge base serves as the genie's memory and provides the following capabilities:

• Stores any data and information that is crucial as a contextual reference for the genie to perform its role.
• Can be updated in real-time to ensure the genie always has the most current information.
• Can access metadata to filter by attributes such as created date, source, or knowledge base ID.

How do I create a knowledge base?

Complete the following steps to create a knowledge base:

1

Go to AI Hub > Genies and create or edit a genie.

2

Add a knowledge base in the Knowledge bases section.

3

Choose your data source (Knowledge recipes or Workato GO data sources).

How do I add a knowledge base to a genie?

Complete the following steps to add a knowledge base to a genie:

1

Go to AI Hub > Genies and select your genie.

2

Locate the Knowledge bases section and click + Add.

3

Search for and select the knowledge base you plan to add.

The knowledge base must be in the same project as the genie.

Can a knowledge base be shared across multiple genies?

Yes. Knowledge bases can be shared across multiple genies within the same project. This allows you to reuse the same knowledge base across different genies without duplication.

What file formats are supported when ingesting documents through knowledge recipes?

The following file formats are supported when using the Store document in a knowledge base action in knowledge recipes:

• PDF (.pdf)
• Microsoft Word (.docx)
• Microsoft Excel (.xlsx)
• Microsoft PowerPoint (.pptx)

When should I use a knowledge base versus a database?

Use knowledge bases for semantic search over unstructured content, such as policies, guides, or descriptions. Use databases (accessed through skills) for exact lookups, counts, and filtered queries over structured data.

Refer to Knowledge bases and databases for more information.

Why isn't my genie finding information in structured data files?

Knowledge bases use semantic search, which struggles with structured formats like JSON or CSV. Raw structured data creates poor search results because semantic search can't effectively parse key-value pairs and repetitive field names.

Solution: Transform structured data into readable prose before adding it to knowledge bases. For example, convert:

• Raw: {"ticket_id": "12345", "status": "open"}
• Prose: "Support Ticket #12345 for Acme Corp - Login timeout issue (Status: Open)"

When to use knowledge bases vs. databases:

• Use knowledge bases for semantic search, such as finding similar issues or discovering patterns.
• Use databases accessed through skills for structured queries, such as counting, filtering, or exact lookups.

Refer to Prepare JSON and API application data for format examples.

Why does my genie return incomplete results for counting or listing questions?

Knowledge bases are designed for semantic search, not comprehensive queries. Each query returns only the 10 most relevant documents. This means that when you ask how many invoices are overdue?, your genie sees only 10 matches even if hundreds exist.

Solution: Use databases accessed through skills for counting, aggregations, or comprehensive lists. Databases query all matching records and return exact counts.

Refer to Knowledge bases versus databases for guidance on when to use each option.

Why isn't my genie finding information in my documents?

Knowledge bases split documents into 8,000-character chunks with no overlap. This means your genie may not find information if a chunk boundary splits mid-concept, or if key content is buried deep in a document.

Quick fixes:

• Structure documents with headings so chunks break at natural divisions rather than mid-paragraph.
• Front-load important information in the first 2,000 characters of documents.
• Keep related concepts together within 8,000-character sections.

Refer to Knowledge base document preparation for more information.

Skills

What are skills?

Skills are tools that enable genies to take actions in your business systems. Skills can query data, create records, send messages, update tickets, or perform any action your Workato recipes can do.

Can skills be shared across multiple genies?

Yes. Skills can be shared across multiple genies within the same project. This allows you to reuse the same skills across different genies without duplication.

Can I publish skill recipes to the community library?

Yes. You can share skill recipes in the community library.

What's the maximum output size for genie skills?

Skills support a maximum 250 KB output. We recommend using a summary skill + detail skill pattern for queries returning large datasets:

• Summary skill: Returns key identifiers and essential fields only, such as IDs, titles, and status.
• Detail skill: Accepts a specific ID and retrieves full details for one record.

The genie presents the summary, then fetches full details only for records the user selects.

Refer to Design skills for databases for more information.

Why is my skill only returning one record instead of the full dataset?

You're likely using Text mode instead of Formula mode in the Return response to Genie action, or you're in Formula mode but selected individual fields instead of the Records datapill. Text mode returns only the last record from a collection.

Solution:

1

Open your skill recipe and locate the Return response to Genie action.

2

Switch to Formula mode.

3

Select the Records list datapill, not individual field datapills.

4

Test to verify all records return.

Refer to Text versus formula mode for more information.

Why does my genie make mistakes with simple math calculations?

LLMs generate text by predicting tokens, not by performing mathematical operations. LLMs can struggle with precise arithmetic, especially multi-step calculations or financial computations.

Solution: Give your genie access to a Python code execution skill that performs calculations accurately. The skill intercepts calculation requests and returns precise results.

When this is critical:

• Financial calculations (expense reports, budgets, invoices)
• Multi-step arithmetic (totals, aggregations, percentages)
• Any scenario requiring precise numerical results

Refer to Troubleshoot arithmetic errors for implementation steps and a pre-built skill template.

Security and authentication

What security features are available in Agent Studio?

Agent Studio provides several security controls:

Access control:

• Use end-user groups to control who can use specific genies.
• Configure Verified User Access (VUA) so skills execute with individual user permissions.

Auditing:

• Genie logs track skill executions and user interactions.
• VUA-enabled skills show which user performed each action in audit trails.

Governance:

• Genies can only execute skills you explicitly add to them.
• Skills respect connected app permissions and access controls.

What is Verified User Access and when should I use it?

Verified User Access (VUA) allows genie skills to execute with each end user's own credentials, rather than the recipe builder's credentials. End users authenticate once when first using a VUA-enabled skill and can manage their connections using the !list_connections keyword in chat.

Connection types:

Skills support two approaches:

User's connection (individual credentials with VUA):

• Actions run with each end user's identity and permissions
• Provides user-level audit trails
• Respects individual access rights in connected apps
• Requires OAuth 2.0 authorization code grant connections

Your connection (the builder's credentials):

• Actions run with the recipe builder's connection
• Works like standard Workato recipes

When to use VUA: Use individual credentials when you need user-level auditing, permission enforcement, or want to eliminate security risks of shared credentials. Use the builder's credentials when individual user permissions aren't required.

Refer to Add verified user access to skill recipes.

Advanced features

What are app events and when should I use them?

App events enable genies to act proactively by responding to triggers from external systems (like Salesforce or Zoom) instead of waiting for users to start conversations.

For example, when a NetSuite access request is approved, an IT Genie can automatically process next steps without user input.

App events automate complex tasks, surface relevant work at the right time, and give genies context earlier in the process.

What is agent orchestration?

Agent orchestration enables genies to work autonomously within recipes. Recipes can assign tasks to genies without user input, and genies can delegate subtasks to other genies in the same project.

When a recipe uses the Assign task to genie action, the recipe job pauses while the genie processes the task autonomously, then resumes when the genie returns its response. This enables complex multi-agent workflows like compliance audits where one genie delegates evidence collection to another specialized genie.

What are the limitations of agent orchestration?

Agent orchestration has the following limitations:

• Genies using verified user access skills can't be used with the Assign task to genie action.
• Genies requiring Business approvals can't process autonomous tasks.
• Genies can only assign tasks to other genies within the same project.

Troubleshooting

Why can't users access or interact with my genie?

Verify the following if users can't access your genie or the genie isn't responding:

• The genie is started: Check status in AI Hub > Genie.
• End user access is configured: Add user groups in AI Hub > Genie > End user access.
• Workato Identity accounts are activated: Users should receive an email invitation.
• The correct user group is assigned to the genie.
• Genie is properly connected to the chat interface.

This applies to genies in Slack, Microsoft Teams, and Workato GO.

Refer to Create a user group for step-by-step instructions.