{
  "name": "CLARA AI",
  "publisher": "MineSpace Inc.",
  "publicPage": "ai.html",
  "workspace": "ai-studio.html",
  "format": "CLARA-BIGRAM-1",
  "model": "Word-bigram model plus a compact trainable neural character model; not a general-purpose LLM",
  "tokenizer": "ASCII English words and selected punctuation",
  "capabilities": [
    "Approved corpus ingestion",
    "Persistent incremental updates",
    "Held-out perplexity",
    "Extractive evidence search",
    "Next-token prediction",
    "Saved read-only agent profiles",
    "Private model export and deletion/rebuild",
    "Neural weight updates and resumable per-document cursors",
    "Neural held-out character perplexity"
  ],
  "limits": {
    "documents": 64,
    "charactersPerDocument": 16000,
    "tokensPerDocument": 8000,
    "agents": 24,
    "documentsPerBatch": 4
  },
  "local": {
    "storage": "browser localStorage; explicit separate local scope",
    "training": "while studio open"
  },
  "hosted": {
    "storage": "private SQLite",
    "authentication": "installation administrator bearer",
    "endpoint": "platform/ai",
    "method": "POST",
    "cron": "optional ai_cron.py; owner must enable automatic processing"
  },
  "dataPolicy": "Only explicitly approved text is ingested. No automatic ingestion of projects, chats, telemetry, enterprise data, private workspaces, or arbitrary websites. Web material must be obtained lawfully and approved with source and license.",
  "agents": "Evidence retrieval, word prediction or neural next-character prediction only; no external tool actions. Profiles share the selected model. Not a general-purpose autonomous LLM.",
  "publicMCP": "clara_ai_spec returns this document only",
  "neuralModel": {
    "format": "CLARA-CHAR-NN-1",
    "parameters": 890,
    "contextCharacters": 4,
    "contextBuckets": 64,
    "embeddingDimensions": 8,
    "alphabetSize": 42,
    "activation": "tanh",
    "objective": "next-character cross-entropy",
    "optimizer": "online SGD",
    "learningRate": 0.065,
    "charactersPerBatch": 2048,
    "heldOutSampleLimit": 2048,
    "normalization": "ASCII lowercase, digits, selected punctuation; other codepoints mapped to space",
    "training": "one streaming pass over approved train documents; persistent per-document cursor and weights",
    "checkpointHash": "SHA-256 of runtime-native serialized neural model; numeric portability is approximate, not bit identity"
  },
  "operations": {
    "status": {},
    "list": {},
    "train": {},
    "tick": {},
    "automatic": {
      "enabled": "boolean"
    },
    "add": {
      "title": "string <= 120 characters",
      "source": "string <= 500 characters",
      "license": "string <= 200 characters",
      "text": "string <= 16000 characters",
      "role": "train | eval",
      "approved": "true"
    },
    "delete": {
      "id": "document id"
    },
    "evaluate": {},
    "agent-save": {
      "name": "string <= 80 characters",
      "goal": "description <= 500 characters",
      "mode": "evidence | predict | neural"
    },
    "agent-list": {},
    "agent-delete": {
      "id": "agent id"
    },
    "agent-run": {
      "id": "agent id",
      "query": "string <= 500 characters"
    },
    "model-export": {}
  }
}
