Skip to main content

Admin Guide

Use this guide to plan, launch, and maintain Siesta AI for an organization. It follows the actual areas admins use in the product: Organization, Users, Teams, Connections, Agents, Workflows, Tool Executions, Audit Log, and Security.

Start in Organization to configure identity, API keys, SSO, security, and Microsoft Entra synchronization. Then create teams, connect tools, review connection token limits where needed, prepare agents and workflows, and invite users. This order keeps access decisions clear before users can run tool-enabled work.

  1. Open Organization and review General, Api Keys, SSO Config, Settings, and Security.
  2. Decide whether users will join through manual invitations, Microsoft Entra ID synchronization, or approved email domains.
  3. Create pilot teams in Teams and assign users from Users.
  4. Add integrations in Connections with Add Integration, then review scopes, functions, shared/private access, and token limits for model connections.
  5. Configure agents with data collections, shared tools, private tools, model settings, and Access visibility.
  6. Build workflows from connection actions and set Access / Sharing to the correct teams.
  7. Review Tool Executions for tool calls and approvals, and Audit Log for organization changes.

First Rollout vs Advanced Operations​

For the first rollout, focus on the minimum setup that lets teams work safely: organization settings, identity, users, teams, shared and private connections, basic production agents, approvals for write-capable tools, Tool Executions, and Audit Log review.

Move advanced work into a later operating phase: connection-specific tuning, MCP or External API payload design, prompt shield internals, eval and feedback loops, detailed analytics interpretation, token economics, and model cost optimization.

Model Selection Defaults​

Choose model defaults by workload, not by hype. A small set of well-understood defaults is easier to govern than giving every agent a different frontier model.

Use this as a starting point:

  • High-stakes reasoning, complex coding, autonomous workflows, and multi-step analysis: consider frontier models 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 where your tenant has access.
  • Production assistants, support agents, summarizers, classifiers, and frequent internal workflows: prefer balanced models such as GPT-5 Mini, GPT-5 Nano, Claude Haiku 4.5, Gemini 3 Flash, Ministral 3, or another approved low-latency model.
  • Long-context and document-heavy agents: check context window, file support, retrieval behavior, and cost before rollout.
  • Voice, transcription, image, or realtime use cases: use the specialized model family configured for that modality instead of a general chat model.

Review model availability, pricing, and provider-specific limits before assigning models broadly. For the current catalog, see Siesta AI Models.

Admin Readiness Checklist​

  • Identity: SSO Config is complete for Microsoft or Google if SSO is used.
  • User source: Entra sync, approved domains, or manual invitations are chosen before broad rollout.
  • Teams: every team has an owner and a clear access boundary.
  • Connections: write-capable tools have confirmation or approval rules where needed.
  • Agents: each production agent has a clear purpose, model, data source, tools, and access setting.
  • Workflows: workflows are tested on non-production records before team sharing.
  • Monitoring: admins know where to inspect Tool Executions, Audit Log, agent feedback, and conversation history.