Skip to main content

Work with Agents

Agents are configured for specific jobs. They may have a selected model, system message, data connection, shared tools, private tools, access settings, interface settings, and prompt evolution history.

Choose the Right Agent​

Pick an agent that matches the work you need done. A sales agent, support agent, analytics agent, and writing agent may all respond differently because they use different instructions and tools.

Before sending sensitive or action-oriented requests, check whether the agent has access to external tools and whether it can make changes outside Siesta AI.

What to Check on an Agent​

  • Overview: quick status, usage, feedback, conversations, and connected data.
  • Configuration: model, system message, data connection, access, shared tools, private tools, and model settings.
  • Interface: public chat URL, web plugin, authenticated chat widget, feedback settings, and upload settings.
  • Prompts: prompt blocks that control behavior.
  • Analysis: usage and token trends.
  • Conversations: previous user interactions with the agent.
  • Feedback: ratings and comments that may explain poor answers.

Understand Outputs​

Agent answers can combine model reasoning, uploaded data, and tool results. If the answer references an external system, verify important results in that system when accuracy matters.

What System Tools Can Do For You​

Some agents have System Tools, which are built-in Siesta AI capabilities rather than ordinary external connections. Depending on the agent configuration, a system tool can:

  • create a task from the current conversation,
  • use web search to verify current information,
  • read a specific URL and summarize visible page content,
  • analyze an uploaded file or spreadsheet,
  • create a downloadable output file.

Ask for these capabilities in plain language. For example, ask the agent to verify with web search, scrape a URL, create a task, or analyze an attached spreadsheet. Admin-only capabilities such as platform configuration, JavaScript execution, and sub-agent orchestration are normally handled by admins or power users.

Ask for Confirmation​

For actions such as sending messages, updating records, or creating tickets, ask the agent to prepare the action first and wait for your approval before executing it.

Useful confirmation wording: Show me the exact Jira ticket / email / Slack message first. Wait for approval before using the tool.

Common Agent Jobs​

Customer Reply Agent​

Use when you need a support-style answer based on approved knowledge. Ask it to use the relevant Memory or Data source, draft the response, and list anything it could not verify.

Sales Operations Agent​

Use when the work touches HubSpot, follow-up emails, deal notes, or reporting. Ask it to prepare updates first and wait before changing CRM data.

Reporting Agent​

Use when you need a structured summary, table, or recurring report. Give the date range, source collection, and expected columns.

Internal Knowledge Agent​

Use when the answer should come from policies, playbooks, onboarding pages, or company documentation stored in Memory.

Before You Trust The Answer​

  • Check whether the agent used the expected Data or Memory source.
  • If a tool was used, review Tool Executions.
  • If the answer affects a customer, contract, ticket, campaign, or calendar, verify the result in the target system.
  • If the answer seems off, ask: Which sources and tools did you use? What could not be verified?