Skip to main content

Use Chat Effectively

Chat is best for interactive work: asking questions, drafting content, analyzing information, and refining outputs. Good instructions help Siesta AI understand the goal, context, constraints, and expected format.

Work With Text and Files​

You can use chat by typing instructions directly, attaching files, or dragging and dropping files into the conversation. Attach a file when the agent should work from a specific document, image, spreadsheet, report export, screenshot, or other source that is not already connected through Data, Memory, or a tool.

Tell the agent exactly what to do with each file. For example: summarize this PDF for a manager, compare these two contracts, extract action items from these meeting notes, read the attached screenshot, or clean this CSV and return a downloadable version.

If the selected agent has Code interpreter enabled, it can handle more advanced file operations, such as analyzing spreadsheets, transforming CSV data, creating charts, generating cleaned files, or preparing downloadable outputs. Use this for practical file work, but review the result before sending it to customers, importing it into another system, or using it for a business-critical decision.

Write Better Requests​

Include:

  • what you want to achieve,
  • the context the answer should use,
  • any constraints or audience details,
  • the format you want back,
  • whether the answer should be brief or detailed.

Product-Specific Examples​

Good chat requests include the system or source you want to use:

  • Summarize the last customer feedback notes from the Support Memory collection and list the top 5 product issues.
  • Draft a Jira ticket from this conversation. Show me the summary and acceptance criteria before creating it.
  • Compare these uploaded files and return a table with owner, due date, and risk.
  • Use the Sales Operations agent and HubSpot data to prepare a follow-up email, but do not send it yet.

Refine Instead of Restarting​

If the first answer is close but not right, continue the same conversation. Ask for a shorter version, a table, a different tone, more detail, or a specific next step.

When Chat Is Not Enough​

Use an agent when work depends on specific tools or repeatable behavior. Use a workflow when the same sequence of steps should run consistently. Use a task when you want to track work beyond the current conversation.

If a tool action is involved, ask for a preview first. For example: Prepare the Slack message and wait for my confirmation before sending.

Prompt Patterns That Work Well​

Summarize a Source​

Use [source name] and summarize it for [audience].
Return: key points, risks, decisions, owners, and open questions.
If the source does not contain an answer, say what is missing.

Prepare an External Action​

Prepare a [Jira ticket / email / Slack message / calendar event] from this conversation.
Show me the exact content first.
Do not create or send anything until I approve it.

Compare Files or Data​

Compare these files and return a table with:
item, source, difference, risk, recommended next step.
Use only the uploaded files.

Analyze an Uploaded File​

Analyze the attached spreadsheet.
Clean inconsistent values, summarize the main issues, and create a downloadable cleaned version.
Before changing assumptions, tell me what you found.

Continue Work Later​

Create a task from this conversation.
Summary: [short summary].
Prompt: [what the agent should do next].
Set it to In review after the first output.

Use System Tools Safely​

Use web search to verify this, then cite what you found.
Do not rely only on memory.
Scrape this URL and summarize only facts visible on the page.
Create a task from this conversation with summary, prompt, and status Todo.
Analyze this spreadsheet and create a downloadable cleaned version.

Common Chat Mistakes​

  • Asking analyze this without saying which source should be used.
  • Asking an agent to send or create something without requesting a preview.
  • Starting a new conversation when the previous thread has the important context.
  • Mixing unrelated goals in one long conversation.