OpenAI-compatible.
India-built.

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.NEUROCORTEX_KEY,        // neurocortex_v3_live_…
  baseURL: "https://bharatneurotech.com/api/public/v1",
});

const out = await client.chat.completions.create({
  model: "neurocortex-v3",
  messages: [{ role: "user", content: "audit this clause" }],
  max_tokens: 16384,                           // model-card max output
  // Opt-in NeuroCortex modes (all default OFF for low latency):
  neurocortex: {
    deep_thinking: false,
    agent_mode:    false,
    web_access:    false,
  },
});

API is purely metered — every billable call is 2× the equivalent Lab action on Hobby. Builder 1.5×, Scale 1.2×, Sovereign 1.0×. Account and catalogue reads are metered at zero. Opt-in modes add small per-call surcharges before the tier multiplier.

• Pre-flight wallet check — low balance returns 402 before the model is ever called. No 30s waits on a doomed call.
• Default max_tokens 16384 — the engine model-card maximum, so chat & codemaster scaffolds never length-truncate. Sub-256 is rejected at 400.
• Empty responses auto-refund — 502 empty_response_increase_max_tokens with a fix.
• Modes explicitly OFF by default — no surprise tool-call hallucinations, no surprise reasoning latency.
• Hard timeouts — 30s standard, 60s with deep_thinking, 240s for streams.