Skip to main content

​
Core Concepts

​
Workspaces

A workspace is the top-level container for your organization. Every resource — agents, connectors, projects, runs — belongs to a workspace.
  • Each user can belong to multiple workspaces
  • Workspace members share access to all resources within it
  • Billing and subscriptions are managed per workspace

​
Projects

Projects are logical boundaries within a workspace for organizing work. Data is fully isolated between projects:
  • Agents, connectors, cron jobs, chat conversations, and runs are scoped to a project
  • Switching projects updates all views to show only that project’s data
  • A resource with no project belongs to the workspace globally

​
Agents

Agents are AI-powered assistants configured with:
  • System Prompt — Instructions that define the agent’s behavior
  • Provider Key — The AI model and API key used for inference
  • Connectors — External tools the agent can access during task execution
Agents can be used in chat conversations, scheduled via cron jobs, or triggered through tasks.

​
Connectors

Connectors integrate Sontairo with external services. They support:
  • OAuth 2.0 — Secure token-based authentication (Google, Slack, GitHub, etc.)
  • API Keys — Direct key-based access for services without OAuth
  • MCP (Model Context Protocol) — Custom tool servers for advanced integrations
Each connector belongs to a workspace and optionally a project.

​
Tasks & Runs

  • Task — A one-off instruction sent to an agent, optionally with connectors attached
  • Run — The execution record of a task, including status, logs, and output
Runs are tracked in the Runs timeline with detailed step-by-step visibility.

​
Cron Jobs

Schedule agents to run tasks automatically on a recurring basis using cron expressions.
  • Configure schedule, timezone, and prompt
  • Toggle jobs on/off without deleting them
  • Each execution creates a new run

​
Chat Conversations

Interactive sessions with AI agents featuring:
  • Streaming responses
  • Persistent conversation history
  • Agent memory across conversations
  • Tool use during chat (via connected tools)

​
Approvals

Human-in-the-loop workflows that pause agent execution until a user approves or rejects the proposed action. Approvals are workspace-scoped.