We build WFGY, an open-source reasoning and debugging ecosystem for AI systems.
The strongest practical entry point is now the AI Troubleshooting Atlas.One lineage, multiple public entry points, one growing ecosystem under the MIT License.
- ⭐️ AI Troubleshooting Atlas : the main practical entry for broken RAG, agent, and AI workflows.
- ⭐️ Global Debug Card : image-first triage for a single failing run.
- ⭐️ Ecosystem Map : canonical map of how the public WFGY system fits together.
- ⭐️ WFGY 3.0 : frontier reasoning and long-horizon evaluation surface.
If WFGY helps your workflow or thinking, a ⭐ on the repo helps more people discover it.
WFGY is for people who need structured debugging and serious reasoning, not just another prompt recipe.
- RAG and agent teams : pipelines run, infra looks healthy, but answers are still wrong or unstable.
- Infra and platform owners : you need a way to inspect and route failures across models, tenants, or deployments.
- Researchers and evaluation teams : you study reasoning, robustness, safety, or stress tests and want concrete observables.
- Founders, PMs, and domain experts : you carry difficult AI workflows and want a more structured system for treating them.
If you do not fit neatly into any of the above, start with the Atlas or the Global Debug Card.
- ⭐️ AI Troubleshooting Atlas : route-first troubleshooting surface for real AI failures.
- ⭐️ WFGY RAG 16 Problem Map : classic 16-problem checklist for broken RAG flows.
- ⭐️ WFGY Global Debug Card : image-first triage layer for diagnosing a single failing run.
- ⭐️ WFGY 1.0 : earliest public conceptual foundation.
- ⭐️ WFGY 2.0 : current reasoning and diagnostic kernel.
- ⭐️ WFGY 3.0 : frontier TXT-based reasoning and evaluation surface.
WFGY is not a single page, a single chart, or a single claim.
The public system is easiest to read as:
- one version lineage : WFGY 1.0 → WFGY 2.0 → WFGY 3.0
- one strong practical wedge : AI Troubleshooting Atlas, Problem Map, Global Debug Card, and Global Fix Map
- one wider application and evaluation surface : TXTOS, related modules, and WFGY 3.0
- one public proof and collaboration layer : Adopters, Case Evidence, Recognition Map, Evidence Timeline, Work with WFGY, and Support
The goal is simple: make reasoning failures more visible, reproducible, and fixable.
- ⭐️ Adopters : shortest public adoption summary.
- ⭐️ Case Evidence : what those integrations imply in real systems.
- ⭐️ Recognition Map : broader ecosystem record of integrations, citations, curated lists, and public mentions.
- ⭐️ Evidence Timeline : historical timeline of how WFGY became public, usable, and externally legible.
Most current public references point first to the WFGY ProblemMap / 16-problem failure checklist line.
A smaller but growing set also uses WFGY 3.0 as a long-horizon, TXT-based reasoning and evaluation surface.
This does not mean every project uses the full WFGY ecosystem.
In many cases, WFGY appears first as a practical diagnostic layer for RAG and agent pipelines.
Representative integrations
| Project | Stars | Segment | How it uses WFGY ProblemMap | Proof (PR / doc) |
|---|---|---|---|---|
| LlamaIndex |  |
Mainstream RAG infra | Integrates the WFGY 16-problem RAG failure checklist into its official RAG troubleshooting docs as a structured failure mode reference. | PR #20760 |
| RAGFlow |  |
Mainstream RAG engine | Introduced a RAG failure modes checklist guide to the RAGFlow documentation via PR, adapted from the WFGY 16-problem failure map for step-by-step RAG pipeline diagnostics. | PR #13204 |
| FlashRAG (RUC NLPIR Lab) |  |
Academic lab / RAG research toolkit | Adapts the WFGY ProblemMap as a structured RAG failure checklist in its documentation. The 16-mode taxonomy is cited to support reproducible debugging and systematic failure-mode reasoning for RAG experiments. | PR #224 |
| DeepAgent (RUC NLPIR Lab) |  |
Academic lab / agent research | Adds a multi-tool agent failure modes troubleshooting note inspired by WFGY-style debugging concepts for diagnosing tool selection loops, tool misuse, and multi-tool workflow failures in agent pipelines. | PR #15 |
| ToolUniverse (Harvard MIMS Lab) |  |
Academic lab / tools | Provides a WFGY_triage_llm_rag_failure tool that wraps the 16 mode map for incident triage. |
PR #75 |
| Rankify (University of Innsbruck) |  |
Academic lab / system | Uses the 16 failure patterns in RAG and re-ranking troubleshooting docs. | PR #76 |
| Multimodal RAG Survey (QCRI LLM Lab) |  |
Academic lab / survey | Cites WFGY as a practical diagnostic resource for multimodal RAG. | PR #4 |
| LightAgent |  |
Agent framework | Incorporates WFGY ProblemMap concepts into its documentation via a Multi-agent troubleshooting (failure map) section, providing a structured symptom → failure-mode → debugging checklist for diagnosing role drift, cross-agent memory issues, and coordination failures in multi-agent systems. | PR #24 |
If you maintain an AI system, research project, or infra platform and want to explore collaboration around WFGY, start here:
- ⭐️ Work with WFGY : entry point for pilots, audits, and structured collaboration.
- ⭐️ Pilot Offer One-Pager : compact view of what a WFGY pilot can look like.
- ⭐️ Sample Deliverable : sample structure of a WFGY pilot return package.
- ⭐️ Support WFGY : support the continued development of the public ecosystem.
You can also:
- open an issue describing your current system and failure modes
- reference the matching WFGY ProblemMap number if you already know it
- reach out via Discord for exploratory discussion
We are especially interested in:
- RAG or agent teams testing WFGY diagnostics in production-like settings
- research groups designing stress tests or observables on the atlas line
- platform owners exposing WFGY-style diagnostics to their users
The long-term goal is simple: make reasoning and debugging layers a normal, visible part of AI systems.
If this ecosystem is useful to you, a ⭐ helps more people discover it.