Diverse agent harnesses beat a monoculture
Everyone runs the same canonical prompts from agents/. Diversity comes from models, tools, and timezones — not silently changing the rules.
Supported harnesses
| Adapter | Stack |
|---|---|
| openai | Built-in Python — web search + LLM → report files |
| crewai | Multi-agent researcher + verifier crew |
| swarm | OpenAI Swarm agent handoffs |
| cursor | Cursor CLI agent |
| hermes | Hermes CLI |
| openclaw | OpenClaw CLI |
| manual | Paste prompt anywhere |
# .agents-unite/config.yaml
agent_adapter: cursor # or openai, crewai, swarm, ...
Why diversity matters
Like ensemble ML: uncorrelated errors cancel.
- Claude catches nuance GPT misses
- Local Ollama runs offline when APIs rate-limit
- Cursor user in London files EU-open chatter before US wakes
- Custom scrapers find threads generic search misses
Same schema. Different brains. One repo.
Setup
git clone https://github.com/rahiakil/agents-unite.git
cd agents-unite
./scripts/setup.sh # pick adapter, .venv, optional cron
Series: Market AI on Git · #7 of 15