nomicore-ai/nomicore

Nomicore

English | 中文

The database built for agents.

面向 Agent 的数据库

Example: when revenue: 120 is not enough

Suppose an Agent receives this result from a conventional database:

{
  "month": "2025-01",
  "revenue": 120
}

The value looks simple, but the Agent cannot safely use it without asking more questions:

• Is revenue measured in dollars, thousands of dollars, or another currency?
• Is it recognized revenue, invoiced revenue, or cash collected?
• Does it include tax, refunds, and intercompany transactions?
• Which schema version produced this record?
• Did the definition change between this month and historical records?

With Nomicore, the result includes both the data and the information needed to interpret it:

{
  ok: true,
  value: {
    month: '2025-01',
    revenue: 120
  },
  schema: `# readData []

{
  month: Pattern<"^[0-9]{4}-(0[1-9]|1[0-2])$"> // Reporting month in YYYY-MM format
  revenue: Range<0, 999999999> // Recognized revenue in USD thousands, excluding tax and refunds; accounting policy 2025-v2
}
`,
  truncated: false
}

Pattern<"…"> means the value must match the specified format. Range<0, 999999999> means the value must be a number within that range. The comments explain what each field means and how it should be interpreted.

The Agent now knows that 120 means USD 120,000 of recognized revenue under accounting policy 2025-v2. If an older record uses a different shape or definition, that record can retain its own schema and semantics rather than being silently interpreted under the latest rules.

When this result is sent to another Agent, its schema and semantics travel with it. The receiving Agent does not need access to a separate data dictionary or undocumented organizational context before it can interpret the value correctly.

Why Nomicore

Why traditional databases fall short in the Agent era

Traditional databases were primarily designed for applications written by humans. An application can encode data structures, business rules, and error handling in advance, but an Agent works on dynamic tasks and must understand data and its boundaries while reading, changing, sharing, and monitoring it. Traditional databases leave important gaps at every step: