API Documentation

Memory Arena provides a public JSON API for programmatic access to all memory system scores, benchmark metadata, and architecture taxonomy data. All endpoints return static JSON files and require no authentication.

Base URL

https://memory-arena.vercel.app

Endpoints

Method	Path	Description
GET	/api/v1/systems.json	Returns all memory systems with their scores, capabilities, and metadata.
GET	/api/v1/benchmarks.json	Returns all benchmark metadata including measures, evaluation type, and categories.
GET	/api/v1/taxonomy.json	Returns all 9 architecture families with descriptions, strengths, and weaknesses.
GET	/api/v1/systems/[slug].json	Returns a single memory system by its slug identifier.
GET	/api/v1/benchmarks/[slug].json	Returns a single benchmark by its slug identifier.

Example Responses

GET/api/v1/systems.json

Returns all memory systems with their scores, capabilities, and metadata.

[
  {
    "id": "mem0",
    "name": "Mem0",
    "organization": "Mem0 Inc.",
    "released": "2024-08",
    "architecture": {
      "family": "hybrid",
      "variant": "Vector + Graph + Episodic",
      "details": "..."
    },
    "domain": ["agent-memory", "personalization"],
    "scores": { "longmemeval": { "overall": 68.5, "source": "...", "date": "2025-09" } },
    "capabilities": { "retrievalAccuracy": 0.86, ... }
  }
]

GET/api/v1/benchmarks.json

Returns all benchmark metadata including measures, evaluation type, and categories.

[
  {
    "id": "longmemeval",
    "name": "LongMemEval",
    "fullName": "LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory",
    "publisher": "Salesforce / UCSD",
    "venue": "arXiv 2024",
    "evaluationType": "automatic",
    "testPrompts": 500,
    "dimensions": 5,
    "measures": ["Information extraction", ...],
    "doesNotMeasure": ["Real-time latency", ...],
    "lowerIsBetter": false,
    "category": "cross-session-memory"
  }
]

GET/api/v1/taxonomy.json

Returns all 9 architecture families with descriptions, strengths, and weaknesses.

[
  {
    "id": "vector-rag",
    "name": "Vector RAG",
    "description": "Dense embedding stores queried by approximate nearest-neighbor search...",
    "keyModels": ["pinecone", "weaviate", ...],
    "strengths": ["Mature tooling", ...],
    "weaknesses": ["Recall degrades on multi-hop", ...]
  }
]

GET/api/v1/systems/[slug].json

Returns a single memory system by its slug identifier.

{
  "id": "letta",
  "name": "Letta",
  "organization": "Letta (formerly MemGPT)",
  "released": "2024-09",
  "architecture": { "family": "agentic-workflow", ... },
  "scores": { ... },
  "capabilities": { ... }
}

GET/api/v1/benchmarks/[slug].json

Returns a single benchmark by its slug identifier.

{
  "id": "longmemeval",
  "name": "LongMemEval",
  "fullName": "...",
  "publisher": "Salesforce / UCSD",
  "venue": "arXiv 2024",
  "measures": [...],
  "doesNotMeasure": [...]
}

V2 API (Coming Soon)

The current API serves static JSON files generated at build time. V2 will introduce a full REST API with filtering, pagination, sorting, and real-time updates. Subscribe to the GitHub repository for updates.