Rifft — cross-framework debugger for multi-agent AI systems.
Rifft helps developers debug multi-agent AI systems by making agent decisions, message flow, tool calls, and failure cascades visible across frameworks. It is intentionally opinionated: Rifft is for debugging multi-agent behavior, not for generic LLM observability.
Add a real communication-graph screenshot or GIF here before public launch.
- Cross-framework trace ingestion for multi-agent runs
- Agent-to-agent graph visualization
- Timeline and per-agent debugging views
- MAST failure classification
- Self-hosted deployment with Docker Compose
- Prompt management UI
- LLM evaluation or scoring
- AI gateway or proxy features
- General APM or application monitoring
- Single-LLM call tracing as the primary feature
| Framework | Status |
|---|---|
| CrewAI | Full |
| AutoGen / AG2 | Full |
| MCP | Full |
| LangGraph | Planned |
| Custom agents via SDK | Full |
pip install rifft rifft-crewaiimport rifft
import rifft.adapters.crewai
rifft.init(project_id="my-project", endpoint="http://localhost:4318")
# Your existing crew code unchanged
crew = Crew(agents=[...], tasks=[...])
result = crew.kickoff()
# Open http://localhost:3000 to see the tracegit clone https://github.com/nedbpowell/rifft.git
cd rifft
cp .env.example .env
docker compose up -d --build
open http://localhost:3000Default local endpoints:
- Web UI:
http://localhost:3000 - API:
http://localhost:4000 - Collector HTTP:
http://localhost:4318 - Collector gRPC placeholder:
localhost:4317
rifft/
├── apps/
│ ├── api
│ └── web
├── infra/
│ └── docker
├── packages/
│ ├── adapters/
│ │ ├── autogen
│ │ ├── crewai
│ │ └── mcp
│ ├── collector
│ ├── sdk-js
│ └── sdk-python
└── docs/
- Copy
.env.exampleto.envfor local development. - Current implementation tracker: docs/implementation-tracker.md
- Current Phase 0/1 plan: docs/phase-0-1-plan.md
- Use scripts/cleanup-demo-traces.sh to remove stale pre-fix demo traces from local storage.
- Architecture notes: docs/architecture.md
- Docs/wiki placeholder: GitHub Wiki
- Contributing: CONTRIBUTING.md
MIT