Sazabi
ChatDelegation

Overview

Complex investigations can spawn subagents and run autonomously in the background.

For complex investigations that require exploring multiple paths or running autonomously, the assistant supports delegation. This includes both inline subagent delegation and fully autonomous background investigations.

What is delegation?

Delegation allows the assistant to break down complex problems and work more efficiently:

  • Subagents: Child investigations that run inline within your conversation, focusing on specific aspects of a larger problem
  • Background investigations: Autonomous monitoring that runs on a schedule, inspecting your systems and alerting you to problems

Both approaches extend the assistant's capabilities beyond single-turn conversations.

Delegation options

When to use each

ScenarioUse subagentsUse background
Multi-part question in conversationYesNo
Autonomous monitoringNoYes
Parallel hypothesis explorationYesNo
Continuous system health checksNoYes
Deep dive into a specific aspectYesNo
Proactive alert generationNoYes

How delegation works

Subagent workflow

When you ask a complex question, the assistant may:

  1. Identify that the question has multiple independent parts
  2. Spawn one or more subagents to investigate each part
  3. Collect findings from all subagents
  4. Synthesize results into a comprehensive answer

You see this happening in real time as nested tool calls in your conversation.

Background workflow

When background investigations are enabled:

  1. A scheduled heartbeat triggers every 30 minutes
  2. The assistant inspects logs and system state
  3. If problems are found, alerts are created
  4. Alerts are delivered through your configured channels

Background investigations run in hidden threads that only surface when they find something worth alerting.

Capabilities and limits

Both subagents and background investigations share most of the assistant's investigation tools, with some restrictions:

Can do:

  • Query logs from all connected backends
  • Run code in the sandbox
  • Search the web for documentation
  • Create artifacts (charts, tables, diagrams)

Subagent limits:

  • Maximum 2 levels of nesting (main -> subagent -> nested subagent)
  • Cannot trigger alerts or modify settings
  • Results must fit within response limits

Background limits:

  • Fixed 30-minute schedule
  • Resource caps per run
  • Alerts require corroborated evidence

These limits ensure delegation remains efficient and does not overwhelm your systems or generate noise.

Getting started

To start using delegation:

  1. Subagents: Just ask complex questions. The assistant automatically determines when delegation would help and spawns subagents as needed.

  2. Background investigations: Enable the feature in your project settings. See Background Investigations for setup instructions.