jsonfirst-ai-agents

AI agents fail when they act on ambiguous input.

JSONFIRST adds a structured intent governance layer before any agent executes — parse, validate, and route natural language deterministically.

Intent → Govern → Execute.

JSONFIRST helps build reliable AI agents by converting natural language into structured JSON intent.

User input (any language, any format)
    ↓
JSONFIRST (parse + govern)
    ↓
JDON: { action, object, confidence, domain }
    ↓
canExecute() → true/false
    ↓
Agent execution (safe)

Quick Example

const { parseIntent, canExecute, getAction } = require('jsonfirst-ai-agents');

const jdon = await parseIntent(
  "cancel the premium subscription for user 42",
  process.env.JSONFIRST_API_KEY
);

if (canExecute(jdon, 0.85)) {
  const action = getAction(jdon); // → "cancel"
  await agents[action].run(jdon);
} else {
  console.log('Low confidence — requesting clarification');
}

Why add a governance layer

Without JSONFIRST:

User: "remove john" → agent deletes user → irreversible

With JSONFIRST:

User: "remove john" → JDON: { action: "delete", confidence: 0.91 }
→ canExecute(jdon, 0.9) → true → agent deletes safely

Installation

npm install jsonfirst-ai-agents

Get your API key at jsonfirst.com → Dashboard → API Console.

API

const { parseIntent, canExecute, getAction } = require('jsonfirst-ai-agents');

| Function | Description | |---|---| | parseIntent(text, apiKey, opts) | Returns structured JDON | | canExecute(jdon, threshold) | Returns true if confidence ≥ threshold (default: 0.8) | | getAction(jdon) | Returns normalized action string |

Governance Modes

| Mode | Use case | |---|---| | ANTI_CREDIT_WASTE_V2 | Default — fast, token-efficient | | MAX_PERFORMANCE | Maximum accuracy for critical actions | | STRICT_PROTOCOL | Enforces full JSONFIRST compliance | | GUARDIAN_MODE | Enhanced safety for destructive actions |

Links

• JSONFIRST
• Documentation
• npm

License

MIT © JSONFIRST