Quartz 4

Agents Tier Structure

2 min read

Agents Tier Structure

This map shows all 44 active agents organized by compute tier (Strategic / Operations / Automation / Special), plus the 3 formally retired shells. Read this before assigning a task to an agent, modifying agent SOUL.md files, or understanding cost allocation by tier.

Diagram

graph TD
    T1["Tier 1 — Strategic\nOpus 4 · 2 agents"]
    T2["Tier 2 — Operations\nSonnet 4 · 13 agents"]
    T3["Tier 3 — Automation\nHaiku 4.5 · 18+ agents"]
    SP["Special / Mixed Model\n3 agents"]
    RT["Retired shells\n3 archived"]

    T1 --> aurora["aurora\nChief AI Ops, Discord"]
    T1 --> solara["solara\nStrategy, governance"]

    T2 --> main["main — Orchestration, cron"]
    T2 --> atlas["atlas — Architecture, IL ops"]
    T2 --> acquisitions["acquisitions — Inbound leads"]
    T2 --> dispo["dispo — Disposition mgmt"]
    T2 --> betterfiles["betterfiles — Doc mgmt, TC"]
    T2 --> backfill["backfill — Data ETL"]
    T2 --> ops_t2["bizintel · reri · ryan-norcal\ndavid-socal · hhill-socal\nauctionscout · prediction-trader"]

    T3 --> auto_web["research · ops · mktg\nwebops · social"]
    T3 --> auto_real["bettershowings · betterhomeoffers\nbettercredit · betterarbitrage\ntradelines · crowdfunding · betterbets"]
    T3 --> auto_fund["fund-calls\n→ angel-calls · h2-calls · h3-calls"]
    T3 --> auto_il["investorlift\n→ il-angel · il-h2 · il-h3"]
    T3 --> auto_inactive["bettertrading · tablegames\n(inactive)"]

    SP --> chatgpt["chatgpt — GPT-4o"]
    SP --> kimi["kimi — Moonshot"]
    SP --> anthropic_cfg["anthropic — config only"]

    RT --> retired["stefania · stefania-asst\ndefault (archived)"]

How to read this

  • Tier 1 (Strategic) contains exactly 2 agents: aurora and solara. These use Opus 4 and should be invoked only for high-complexity, high-value tasks.
  • Tier 2 (Operations) has 13 active agents (betteracq and dealcmd are merged/retired). These use Sonnet 4 and handle most domain-specific workflows.
  • Tier 3 (Automation) runs Haiku 4.5 — fast, cheap, high-volume. fund-calls is a parent that orchestrates angel-calls, h2-calls, h3-calls. investorlift orchestrates its three sub-agents.
  • Special agents use non-Anthropic models (GPT-4o for chatgpt, Moonshot for kimi). anthropic is a config-only entry, not a real agent.
  • 3 retired shells (stefania, stefania-asst, default) remain in _retired/ per feedback_archive_not_delete.md — do not delete.

See also

  • ARCHITECTURE.md — Sections 2.2–2.5 (agent tables with Discord channel bindings)
  • AGENT-REGISTRY.md — domain ownership per agent
  • CLAUDE.md — Agent Identity First mandatory protocol before changing any agent

Graph View

Backlinks