---
created: 2026-03-12
source: Rivet
tags: [agent-archive, rivet]
---

# 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

| Agent | Role | Primary Model | Provider | Fallback | Status |
|-------|------|--------------|----------|----------|--------|
| Rivet | Chief of Staff | Claude Opus 4.6 | Anthropic | GPT-5.2 | ✅ No change |
| Builder | Coding | Claude Opus 4.6 | Anthropic | Claude Sonnet 4 | ✅ No change (Claude Code locked) |
| Susan | Sales/Marketing | Gemini 3.1 Pro | Google | Claude Sonnet 4 | ⏳ Need Gemini API setup |
| Harper | Finance/Legal | GPT-5.2 | OpenAI | Claude Sonnet 4 | ⏳ Pending |
| Herald | Communications | Grok 4.1 | xAI | Gemini 3.1 Pro | ⏳ Need xAI API setup |
| Sentinel | DevOps | DeepSeek R1 | DeepSeek | Kimi K2.5 | ⏳ Pending |
| Radar | Intelligence | Kimi K2.5 | Moonshot | DeepSeek R1 | ⏳ Pending |
| Cog | Operations | MiniMax M2.5 | MiniMax | Kimi K2.5 | 🔄 FIRST — testing now |

## Rollout Order
1. **Cog** → MiniMax M2.5 (lowest risk, routine ops)
2. Radar → Kimi K2.5 (already free, easy swap)
3. Sentinel → DeepSeek R1 (already have access)
4. Harper → GPT-5.2 (already have API key)
5. Susan → Gemini 3.1 Pro (need to configure)
6. Herald → Grok 4.1 (need xAI API access)

## Notes
- Builder stays Anthropic — Claude Code is provider-locked
- Rivet stays Opus — hardest coordination job, dirty data processing
- Each fallback is from a DIFFERENT provider than primary
- 7 different AI companies across 8 agents