The only AgentX agent with 3-Layer Fallback and guaranteed response.
A2A-compatible | AgentBeats Validation Framework
| Feature | Purple |
|---|---|
| 3 Fallback Layers | Backend → LiteLLM (multi-provider) → Heuristic |
| Guaranteed Response | Heuristic always provides orientation |
| Multi-Provider | OpenAI, Anthropic, OpenRouter – sequential fallback |
| 13 Curated Debate Topics | AI, Climate, Privacy, Healthcare, Education, Immigration, MinWage, Vaccines … |
| Structured JSON Output | Artifact "Structured" for evaluators (thesis, arguments, conclusion) |
| 6 Skills, 4–6 Examples/Skill | Green-Agent-Matching for AgentBeats |
| Health Endpoint | /health for observability |
→ Details: ARCHITECTURE.md
src/
├─ server.py # Server setup and agent card configuration
├─ executor.py # A2A request handling
├─ agent.py # Debate logic, topic matching, structured output
└─ messenger.py # Response formatting and artifact generation
scripts/ # Benchmark, analysis, and leaderboard tools
tests/ # Pytest test suite
docker compose up --buildpip install -e .
python -m src.servercurl http://localhost:8080/health| Metric | Result |
|---|---|
| AgentBeats Win Rate | 96.5% (201/208) |
| Fallback Layers | 3 (Backend → LiteLLM → Heuristic) |
| Debate Topics | 13 curated |
| Response Guarantee | 100% |
Purple uses a 3-layer fallback architecture to guarantee responses:
- Layer 1 – Backend: Direct LLM call via configured provider
- Layer 2 – LiteLLM: Multi-provider fallback (OpenAI → Anthropic → OpenRouter)
- Layer 3 – Heuristic: Rule-based response generation (always available)
→ See ARCHITECTURE.md for the full technical breakdown.
pytest tests/ -vOr run the full debate benchmark:
python scripts/agentx_purple_benchmark_suite.py| Document | Description |
|---|---|
| ARCHITECTURE.md | System architecture and fallback layers |
| BEST_DEBATE_STRATEGY.md | Debate strategy and optimization |
| BRILLIANT_DEBATE_SOLUTIONS.md | Battle-tested debate solutions |
| TESTING_CHECKLIST_AGENTX.md | QA and testing checklist |
| BUILD_AND_PUBLISH.md | Build and deployment guide |
- AgentBeats Framework – Debate evaluation platform
- CellRepair AI – Main website
- Demo Video – YouTube demo
Built by Oliver Winkel at CellRepair AI