Sub-Agent Collaboration

Mechanism

The main agent creates sub-agents for sub-tasks:

Independent sessions and contexts per sub-agent
Main agent aggregates results
Right panel “SubAgents” tab shows real-time status

Model Tiering

Settings > Models > Sub-agent tiers:

Tier	Best for	Recommended model
Simple	Research, information gathering, format validation	A cheap model
Standard	Daily coding, document writing, routine analysis	A mid-tier model
Hard	Complex reasoning, architecture design, multi-file refactoring	A strong model

Unconfigured tiers follow the parent session model.

Sub-Agent Discipline

Sub-agent tool calls follow the sub-agent approval policy (Settings → Security → Sub-agent approval)
Follow parent session (default): approval cards show in the main chat’s approval area, tagged with the originating sub-agent, at the same approval level as the parent
With fully automatic selected, routine sub-agent operations are auto-approved (there’s no UI to show approval dialogs), while dangerous operations still always require confirmation

Two guardrails hold in every mode: dangerous operations always require confirmation, and sub-agent deletion is always forbidden.

Seven Sub-Agent Types

A type can be assigned when dispatching, each with its own tool domain and persona:

Type	What it does	What it can touch
explore	Read-only research, small-model tier by default	Read code and search; change nothing
review	Read-only code review	Read code and git history; change nothing
verify	Verification: run tests, build, reproduce issues	Read and execute, but no workspace writes
build	Implementation: write code, edit files	Inherits the parent’s full toolset (cannot spawn sub-agents)
craft	Office output	Docs/sheets/slides/PDF/OCR/transcription + images and video
vision	Visual output	Design canvas + images/video/charts
await	Background watch: monitor a started background task to completion, report once	Watch only, no changes, no new processes

Without a type, the sub-agent inherits the parent’s full toolset. You can also explicitly restrict its tools — permissions never exceed the main session’s.

Types show as badges next to each sub-agent’s name in the panel, so you can see who’s doing what at a glance.

Worktree Isolation and Snapshots

Coding sub-agents run in their own worktrees, so their changes never touch your main directory. Output is snapshot-protected and recoverable even after cleanup. See Git Worktrees.

Agent-to-Agent Collaboration

Sub-agent results flow back to the main session; the main session can also consult experts, and experts can send private messages to each other (with receipts). See Agent Communication.

Example

Do three things simultaneously: 1) Research competitor pricing 2) Organize monthly expenses 3) Write a product update announcement

Three sub-agents execute in parallel; the main agent aggregates.

External Sub-Agents

LanMate can connect external sub-agents (Codex / Claude Code). After configuring the external agent path in settings, the assistant can call them as needed.