Classification methodology
Classification shows what a project does and how central AI is to its core technology. It is not inferred from a token's name, price, or roadmap promises. Rule version: 3.4.
Four dimensions
| Dimension | Question it answers | Example |
|---|---|---|
| Relevance | How important is AI to the project's core function? | Protocol, infrastructure, or peripheral feature |
| Industry | Where is the product used? | Finance, media, robotics |
| Stack | Which technology layer does it provide? | Compute, inference, data |
| Technique | Which approaches are implemented and evidenced? | RAG, agentic systems, ZKML |
Relevance is required. Other tags need evidence. Two to four tags are usually enough, and their count is not a substitute for verification.
How we assign a project role
| Role shown on the coin page | What must be evidenced |
|---|---|
| Core Protocol | Models, data, or agents are central to the protocol's function. |
| AI-Native L1 / Blockchain | The network is designed for AI execution, training, data, or verifiable computation. Partnerships alone do not qualify. |
| Infrastructure | There is a functioning compute, inference, training, data, or tooling layer for AI. A proprietary model is not required. |
| Applied & Agentic Applications | Meaningful first-party AI logic or agents drive the product's core function. |
| AI-Enabled / Peripheral | The main product works without AI, while a model or chat interface is added on top. |
| Non-AI / Remove | No technical link is substantiated. This is an internal exclusion state, not a public segment. |
Controlled vocabularies
Categories are not invented for individual projects. The labels below match those shown on English coin pages.
Industry
- DeFi & Finance
- Gaming & Entertainment
- Enterprise & Business Tools
- Healthcare & Biotech
- Media & Content Generation
- DeSci & Research
- Cybersecurity & Defense
- Social & Creator Platforms
- Robotics & Physical AI
Stack
- Decentralized Compute & GPU DePIN
- Inference & Model Serving
- Model Training & Fine-tuning
- Data Markets & Labeling
- Agents & Autonomous Platforms
- Developer Tools & Middleware
- Indexing, Search & Retrieval (RAG)
- Privacy, ZKML & Secure AI
- Consumer & Application Layer
Technique
- Large Language Models (LLMs)
- Multimodal
- On-chain & Verifiable AI
- RAG & Retrieval-Augmented Generation
- Agentic Systems & Autonomous Agents
- Decentralized & Federated Learning
- Generative
- ZKML & Cryptographic AI
How evidence is reviewed
- SourceStart with documentation, whitepapers, architecture, and GitHub. Blogs and media can add context but do not replace technical evidence.
- FactIdentify implemented functions and map them to the controlled vocabulary. Promises and names are not implementations.
- ReviewLook for contradictions and inflated relevance. When evidence is thin, choose the more conservative role or request human review.
| Borderline case | Decision |
|---|---|
| A general service adds chat using a third-party model | Peripheral feature unless first-party AI logic is central to the product. |
| A GPU network serves documented AI workloads | It may be infrastructure even without a proprietary model. |
| A project calls its bot an “agent” | The agentic-technique tag requires planning and action execution, not just a name. |
| A sensor network claims a robotics link | Direct work with physical systems is needed. Generic DePIN is insufficient. |
What enters the public catalogue
Only published, valid projects contribute to public lists and aggregates. Segment pages with fewer than five published projects receive noindex and are excluded from the sitemap. Permanent descriptions cover purpose and architecture, while changing prices, volumes, and GitHub counts are shown separately. See the data catalog for provenance.
Public sources may be incomplete or outdated. A classification reflects the evidence available at review time and can change. It does not establish project safety or tokenomics quality.