User Guide

Use this guide when you need to finish real work in Siesta AI: ask an agent to research something, draft a customer message, create a Jira ticket, summarize meeting notes, use uploaded files, check why a workflow failed, or understand why a tool is missing.

The pages are organized by the decisions users make during work. Start with Chat for one-off thinking, Agents for role-specific work, Tasks for follow-up, Data or Memory for approved knowledge, and Workflows when the same process should run again.

What Do You Need To Do?

• Ask a question or draft something: use Chat and include source, audience, format, and next action.
• Work through a prepared role: use an Agent that already has the right prompt, model, data, and tools.
• Use company files or policies: ask the agent to use a specific Data collection or Memory page.
• Create or track follow-up work: create a Task, choose the right agent, and move it through Todo, In progress, In review, and Done.
• Run a repeatable process: use a Workflow instead of manually repeating the same steps in chat.
• Connect Gmail, Calendar, Drive, Slack, Jira, or HubSpot: use Connections, then confirm whether the connection is personal or shared.
• Check what happened in another system: open Tool Executions and inspect action, status, arguments, result, and approval state.
• Find previous context: open the original Conversation, recording, task conversation, or agent conversation history.

Decide What To Use First

Start with the smallest feature that matches the job:

1. Use Chat for one-off thinking, drafting, summarizing, or comparing.
2. Use an Agent when the work needs a prepared role, approved data, or specific tools.
3. Use a Task when the work should continue later, move through a status, or be reviewed by someone else.
4. Use a Workflow when the same steps should run repeatedly.
5. Use Data for uploaded files, synced sources, and document collections.
6. Use Memory for maintained knowledge such as policies, playbooks, terminology, and FAQs.

You do not need to understand model connections, retrieval settings, tool binding, scopes, token limits, or workflow node internals before doing useful work. The first layer is simple: choose the right agent or feature, give clear instructions, and check the output before it affects another system.

When Model Choice Matters

Most users should start with the model already configured on the agent. Change models only when the output quality, speed, or cost clearly does not match the job.

As a practical rule:

• Deep reasoning, coding, and complex agent work: use a frontier reasoning model such as GPT-5.2 Pro, GPT-5.2 Thinking, Claude Sonnet 4.6, Claude Opus 4.6, Gemini 3 Deep Think, or Gemini 3.1 Pro when it is available in your organization.
• Daily drafting, summarization, classification, and lower-latency work: use a faster model such as GPT-5 Mini, GPT-5 Nano, Claude Haiku 4.5, Gemini 3 Flash, Ministral 3, or a similar balanced model approved by your admin.
• Large file or long-context work: prefer models with strong long-context support, but still name the exact files, sources, and output format you want.

If you are unsure, keep the agent default and ask an admin which models are approved for your team. For the current model catalog, see Siesta AI Models.

Quick Examples

• Use the Support FAQ Memory collection and draft a reply to this customer. Keep it under 120 words and do not send it.
• Create a Jira ticket from this bug report. Show me summary, description, priority, and acceptance criteria before creating it.
• Use the Q2 customer feedback data collection and list the top 10 issues by frequency in a table.
• Check why the HubSpot to Jira workflow failed and tell me which tool execution needs attention.
• Turn this conversation into a task for the Sales Operations agent and set it to In review.

Which Feature Should I Use?

Need	Use	Why
Explore, draft, compare, rewrite	Chat	Fastest path for one-off work
Work with a prepared role or tools	Agent	The agent already has instructions, data, and allowed tools
Track follow-up	Task	Keeps status, prompt, source conversation, and assigned agent
Ground answers in uploaded files	Data	Uses processed files or data sources
Maintain editable company knowledge	Memory	Good for policies, playbooks, FAQs, and internal pages
Repeat the same multi-step process	Workflow	Runs connected actions in sequence
Debug tool behavior	Tool Executions	Shows action status, inputs, outputs, and approvals