Creating an agent in AgentSpot

Build, test, and share your first AgentSpot agent end-to-end.

Overview

This guide is the full builder walkthrough. Deeper topics — writing instructions, managing sessions, file uploads, and sharing — have their own articles linked at each step.

By the end you’ll have a shared, tested agent your team can use.

Prerequisites

  • You have connected to one or more connectors. While not mandatory, connectors provide more context to your agents and help them act for you, making your agent more valuable.

  • You can answer in one sentence: "What job will this agent do, for whom?"

  • (Recommended) You have 3–5 example questions a real user would ask.

Step 1: Plan the agent

Write these down before opening the builder. Agents with a clear, specific purpose are more useful and more likely to be adopted by your team.

  • Name — short and recognizable (e.g. EMEA Pipeline Assistant).

  • Audience — which team or group.

  • Job — one sentence.

  • Out of scope — what it explicitly will not do.

  • Data it needs — specific sources, tables, or worksheets.

  • Skills it needs — usually none for a first agent.

Agent components

When you create an agent, AgentSpot builds and populates the following components automatically based on your prompt. You can review and adjust each one in the side panel.

  • Connectors: The external applications your agent connects to (for example, Google Calendar or Slack). Added automatically based on your prompt; you can manually add or remove connectors as needed.

  • Tools: Functional capabilities such as web search or image generation. Added automatically; you can manually select or deselect tools as needed.

  • ThoughtSpot models: Your organization’s ThoughtSpot data models. Adding a model allows your agent to work with large structured data sets — a key differentiator for AgentSpot.

  • Skills: Reusable knowledge your agent can draw on (for example, how to format a document using your company template). Some skills are pre-populated; you can add your own.

  • Conversation starters: Suggested prompts shown to users at the start of a chat. Generated automatically; you can edit or add your own.

  • Instructions: The core behavior definition for your agent — its persona, mission, response rules, and guardrails. For example, instructions can prevent the agent from drafting a message or making a tool call without your confirmation. Generated automatically; you can edit the instructions manually at any time.

Tips for writing effective prompts

AgentSpot builds your agent automatically from a single prompt. A clear, specific prompt produces better results.

A strong prompt includes:

  • An objective: What you want the agent to do, in one sentence.

  • Connectors: The applications it should use (for example, "look through my Google Calendar and Slack").

  • Expected output: What a good response looks like.

The following additions improve results further:

  • Examples: Provide examples of past work you want the agent to replicate. This helps the agent understand your preferred format, tone, and output more effectively than step-by-step instructions.

  • Audience: Describing the target audience helps the agent adjust tone, length, and complexity.

Step 2: Create a new agent

  1. On the AgentSpot home page, click Create new agent.

  2. In the prompt field, describe the agent you want to build.

    AgentSpot automatically builds the agent from your description, selecting the appropriate connectors, tools, skills, and generating instructions. The side panel opens as the agent builds, showing each component as it is configured.

  3. Review the components in the side panel and adjust as needed. For details on each component, see Agent components.

  4. Enter a name for your agent and select an avatar.

  5. Click Save agent in the upper-right corner.

After saving, AgentSpot exits the builder and opens your agent with a set of suggested conversation starters. Test a few conversations before sharing — you can return to edit mode at any time and refine the instructions based on how the agent performs.

Step 3: Share the agent

Powerful agents should be shared! Sharing agents allows others in your organization to leverage the skillset of your agent, and prevents “AI silos”, where every contributor creates similar agents that won’t all behave in the exact same way, potentially creating confusion in the organization.

  1. In the upper-right corner, click Share agent.

  2. In the Add people or groups field, find the people or groups you want to share with. You can share with your entire organization by selecting the organization-wide option.

  3. For each person or group, select Can use or Can edit.

  4. Click Done.

The agent now appears in the Agents dashboard for the people and groups you shared it with.

Connecting applications

To provide value, agents require context and the ability to act. If an agent requires external data (e.g., from Notion or Salesforce), you will be prompted to connect.

  1. Identify requirements: The agent interface indicates if an application is not yet connected.

  2. Authenticate: Click Connect.

  3. Authorize: Complete the OAuth flow in the pop-up window to grant AgentSpot access.

  4. Once connected, the agent can search for context and perform tasks such as creating tickets or updating records.

Managing your agents

When you find a useful agent in the library, you must decide how to keep it:

Action Description Best For…​

Save to team

Bookmarks the shared agent. Updates made by the owner are reflected automatically.

Users who want to use the agent as-is and receive future improvements.

Make a copy

Creates a private version of the agent.

Users who want to modify instructions or add private connectors.

Using your agents

Once set up, your agents are available from the AgentSpot home page.

  • Chat history: Resume any previous conversation from the left-hand panel.

  • New session: To start a fresh conversation with an agent, click the new session button in the upper-left corner of the agent window. Start a new session when the conversation context is getting long or when you want to switch to a different topic.

  • Search and filter: Use the list view to find agents by category, such as "Marketing" or "HR."

  • Contextual actions: Agents with connectors (such as Jira or Slack) can perform actions on your behalf, such as drafting and sending messages or creating tickets.

Troubleshooting

Symptom What to try

"I don’t have access to that data."

Confirm the data source is attached and that your admin shared it with your group.

Confident but wrong answers.

Tighten instructions ("only answer from attached data"), lower temperature, add per-table data descriptions.

Formatting instructions are ignored.

Move formatting to the top of the prompt; shorten the overall instructions.

Agent answers out-of-scope questions.

Add an explicit What not to do section naming the topics and re-test.

Citations missing.

Confirm Citations is on in Settings and that the source supports citations. ⚠️ Confirm per-source citation support.
Was this page helpful?
×