# Multi-Model Fleet Plan
*Approved by Michael: 2026-02-25 16:40 AEDT*

## Strategy
Each agent runs a different AI model as primary so every answer comes from a different company with full RateRight context. Maximum diversity of perspective.

## Assignments (Updated 2026-03-12)

| Agent | Role | Primary Model | Provider | Fallback | Fallback Provider | Status |
|-------|------|--------------|----------|----------|-------------------|--------|
| Rivet | Chief of Staff | GPT-5.4 | OpenAI (direct) | DeepSeek-chat | DeepSeek | ✅ Updated |
| Builder | Coding | Claude Opus 4.6 | LocalCC (Max plan) | GPT-5.3 Codex | OpenRouter | ✅ Updated |
| Susan | Sales/Marketing | Kimi K2.5 | Moonshot | DeepSeek-chat | DeepSeek | ✅ Active |
| Harper | Finance/Legal | GPT-5.2 | OpenRouter | MiniMax M2.5 | MiniMax | ✅ Active |
| Sentinel | DevOps | DeepSeek-chat | DeepSeek | Kimi K2.5 | Moonshot | ✅ Active |
| Radar | Intelligence | Grok-3 | OpenRouter/xAI | DeepSeek-chat | DeepSeek | ✅ Active |
| Herald | Communications | Gemini 3.1 Pro | Google (direct) | Kimi K2.5 | Moonshot | ✅ Updated |
| Cog | Operations | MiniMax M2.5 | MiniMax | DeepSeek-chat | DeepSeek | ✅ Active |

## Provider Diversity (no single point of failure)
- **OpenAI direct:** Rivet
- **LocalCC / Max plan:** Builder (Claude Opus 4.6 via claude -p)
- **OpenRouter:** Harper, Radar (2 agents — down from 5)
- **Moonshot:** Susan
- **DeepSeek:** Sentinel
- **Google direct:** Herald
- **MiniMax:** Cog
- **Fallbacks** never use the same provider as primary

## History
- 2026-02-25: Original plan approved (7 providers, Anthropic for Rivet+Builder)
- 2026-03-05: Emergency migration — Anthropic blocked Max plan OAuth, all moved to non-Anthropic
- 2026-03-12: Fixed single point of failure — Rivet→OpenAI direct, Herald→Google direct, diversified all fallbacks away from OpenRouter