Reeve
Configuration

Configuration System

CSS-like config inheritance across universal, org, individual, and workspace layers.

Configuration System

Reeve uses a 4-layer configuration system inspired by CSS specificity. Each layer overrides the one below, giving you smart defaults that can be customized at every level.

The Problem It Solves

Without layers, every agent needs manual configuration — heartbeat templates, tool documentation, safety rules, model defaults. New team members start from scratch. Platform improvements require editing every agent's workspace.

The Layers

Layer 1: Universal        ← Ships with Reeve, every user gets this
  + Layer 2: Org/Team     ← Shared within an organization
    + Layer 3: Individual  ← Personal preferences per user
      + Layer 4: Workspace ← Agent-specific overrides (always wins)
LayerPathWho It's ForExample Content
Universalreeve/config/universal/Every Reeve userRooster Rules, base SOUL.md, safety
Org/data/tenants/{team}/org/Team/companyCompany knowledge, shared tools, team rules
Individual~/.reeve/user/Single userPersonal preferences, writing style
WorkspaceAgent's workspace dirSingle agentAgent-specific instructions, domain context

Resolution: Higher layers win

When Reeve loads config files (AGENTS.md, SOUL.md, TOOLS.md, etc.), it merges them bottom-up. The workspace layer always has the final say.

What Ships in Universal (Layer 1)

Every Reeve agent automatically gets:

  • AGENTS.md — Rooster Rules (safety, memory protocols, protected repos, date/time, formatting)
  • SOUL.md — Base Reeve personality
  • Role templates — Manager, worker, research, assistant templates
  • Model fallback policy — "Never fall below Sonnet for managers"
  • Three-tier architecture rules — Manager/pipeline/worker dispatch pattern
  • Session state protocol — "Write session state after substantive responses"

You never need to copy these into agent workspaces — they're inherited automatically.

Quick Example

A marketing manager agent merges config from all four layers:

Universal AGENTS.md:     Rooster Rules, safety, tool docs
  + Org AGENTS.md:       MindFortress company context, shared knowledge
    + Individual AGENTS.md: Matt's preferences, writing style
      + Workspace AGENTS.md: Marketing-specific instructions, campaign rules

The agent sees all four, merged together. If the workspace defines a rule that conflicts with the org, the workspace wins.

On this page