Reeve
Agent Management

Agent Overview

What Reeve agents are, how roles work, and the difference between built-in and custom agents.

Agent Overview

An agent is a persistent AI identity with its own role, model, workspace, and memory. Agents are the core building block of Reeve — they do the work.

What Makes an Agent

Every agent has:

PropertyWhat it controls
RoleBehavior defaults (model, heartbeat, permissions)
ModelWhich LLM powers it (Claude Opus, Sonnet, GPT-4, etc.)
WorkspaceDirectory with AGENTS.md, SOUL.md, MEMORY.md, and daily logs
Channel bindingsWhich messaging platforms it responds on
HeartbeatPeriodic check-in schedule for autonomous work
Subagent policyWhich other agents it can spawn
Tool policyWhich tools it can use

Roles

Roles provide sensible defaults so you don't have to configure everything from scratch:

RoleModelHeartbeatUse case
coordinatorOpus30 minTop-level orchestrator — routes work, never resets
managerOpus2 hoursDomain lead (marketing, engineering) — spawns workers
workerSonnetNoneTask executor — fast, cheap, resets after 1h idle
researchOpus4 hoursDeep analysis — resets after 24h idle
assistantOpus2 hoursGeneral helper — conversational, resets after 48h idle

Roles are starting points — you can override any default per agent.

Single vs Multi-Agent

Single Agent

The simplest setup: one agent handles everything. Good for personal use, small teams, or getting started.

You ←→ Reeve (assistant)

Multi-Agent

For complex work, agents specialize and coordinate:

Coordinator
  ├── Marketing Manager
  │     ├── Content Writer (worker)
  │     └── Ad Analyst (worker)
  └── Engineering Manager
        ├── Frontend Dev (worker)
        └── Backend Dev (worker)

Managers spawn workers for parallel execution, coordinate results, and report back. Workers are cheap (Sonnet) and reset quickly. Managers and coordinators maintain longer context.

See Multi-Agent for the full coordination system.

Built-in vs Custom Agents

Built-in

Every Reeve installation starts with a default agent (typically assistant role). This agent is ready to go with sensible defaults — you just need an LLM API key.

Custom

Create agents tailored to specific roles:

reeve agents add

The interactive wizard asks for:

  • Name — A human-readable name (e.g., "Marketing Manager")
  • Role — One of the five built-in roles
  • Model — Which LLM to use
  • Workspace — Where to store config and memory

Or create agents from the Cockpit at Settings → Agents → New Agent.

Agent Lifecycle

Create → Configure → Active → [Sessions, Memory, Goals] → Archive/Delete
  1. Create — Via CLI wizard or Cockpit UI
  2. Configure — Set model, tools, heartbeat, workspace files
  3. Active — Agent responds to messages, runs heartbeats, executes goals
  4. Memory grows — Daily logs accumulate, MEMORY.md gets curated
  5. Archive/Delete — When no longer needed

Agents persist across gateway restarts. Their memory, configuration, and workspace survive indefinitely.

Agents are identified by a unique ID (e.g., marketing-manager). This ID is used in configuration, CLI commands, and inter-agent communication.

On this page