NeuroCortex API.
Drop-in OpenAI compatible.

Base URL: https://bharatneurotech.com/api/public/v1. Mint a key in /billing → API Keys. New tokens are issued as neurocortex_v3_live_…. Existing neurocortex_v2_live_… and nck_live_… keys continue to work. The full token is shown once; only its SHA-256 hash is stored.

Authorization: Bearer neurocortex_v3_live_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
Content-Type: application/json

// Node — OpenAI SDK
import OpenAI from "openai";
const client = new OpenAI({
  apiKey: process.env.NEUROCORTEX_KEY,
  baseURL: "https://bharatneurotech.com/api/public/v1",
});

const out = await client.chat.completions.create({
  model: "neurocortex-v3",
  messages: [{ role: "user", content: "Audit this AI policy clause: ..." }],
  max_tokens: 16384,                 // see §5 — model-card maximum
});

# Python — openai
from openai import OpenAI
client = OpenAI(
    api_key=os.environ["NEUROCORTEX_KEY"],
    base_url="https://bharatneurotech.com/api/public/v1",
)
out = client.chat.completions.create(
    model="neurocortex-v3",
    messages=[{"role": "user", "content": "..."}],
    max_tokens=16384,
)

# curl — non-stream
curl -sS https://bharatneurotech.com/api/public/v1/chat/completions \
  -H "Authorization: Bearer $NEUROCORTEX_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "neurocortex-v3",
    "messages": [{"role":"user","content":"One-line summary of ISO 42001."}],
    "max_tokens": 1024
  }'

# curl — streaming (SSE)
curl -N https://bharatneurotech.com/api/public/v1/chat/completions \
  -H "Authorization: Bearer $NEUROCORTEX_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"neurocortex-v3","stream":true,"messages":[{"role":"user","content":"hi"}]}'

# curl — EQCortex (~580B, purpose-built for emotional intelligence)
# Accepts native `system` role. Upstream defaults: temperature=0.6, top_p=0.95, max_tokens up to 32768.
curl -sS https://bharatneurotech.com/api/public/v1/chat/completions \
  -H "Authorization: Bearer $NEUROCORTEX_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "eqcortex",
    "messages": [
      {"role":"system","content":"You are a warm, empathetic writing partner."},
      {"role":"user","content":"Draft a gentle nudge email to a lapsed customer."}
    ]
  }'

# curl — health probe (no auth)
curl -sS https://bharatneurotech.com/api/public/v1/health

neurocortex-codemaster is the code-tuned alias of NeuroCortex — Bharat NeuroTech's rival to GPT Codex, Claude Code, Qwen Code and Google Jules. The server auto-injects a code-only system prompt (refactor / review / scaffold runnable multi-file MVPs), defaults max_tokens to the model-card maximum of 16384, and emits OpenAI-shape responses with model: "neurocortex-codemaster". Try it interactively at /codemaster.

// CodeMaster — scaffold a runnable MVP
const out = await client.chat.completions.create({
  model: "neurocortex-codemaster",
  messages: [{ role: "user", content: "Scaffold a Node + Express RAG chat that calls NeuroCortex." }],
  max_tokens: 16384, // CodeMaster default — model-card maximum output
});

Reasoning, agent tools and web search add latency. They are explicitly disabled on every API call unless you opt in. Set them via the OpenAI-safe extension field neurocortex on the chat body (OpenAI SDKs forward unknown JSON fields untouched):

{
  "model": "neurocortex-v3",
  "messages": [...],
  "max_tokens": 16384,
  "neurocortex": {
    "deep_thinking": false,   // engine called with thinking:false  (≈ 1.5–4× faster)
    "agent_mode":   false,    // tool_choice forced to "none"
    "web_access":   false     // web tool stripped from tool list
  }
}

NeuroCortex is a reasoning-class engine. Setting max_tokens below 1024 is the number-one cause of empty responses — the engine spends its budget on internal reasoning and runs out before emitting the visible answer. The API now defaults to the engine's maximum supported output length so this is no longer a footgun.

• Default if omitted: 16384 (model-card max).
• Minimum accepted: 256 (anything lower → 400 invalid_request).
• Maximum accepted: 16384 (engine cap; requests above are clamped).
• If the engine returns empty content with finish_reason:"length" the API auto-refunds and returns 502 empty_response_increase_max_tokens.

Set stream: true. Response is text/event-stream with data: {…} chunks of object:"chat.completion.chunk" and a terminating data: [DONE]. Cost is debited up-front. If the stream ends with zero visible content deltas, the debit is auto-refunded (upstream failures, empty completions). Reasoning-only frames are suppressed; you never see reasoning_content or <think>…</think> blocks in the output.

Send model: "neurocortex-codemaster" to /v1/chat/completions. The server auto-injects the CodeMaster system prompt (refactor / review / scaffold runnable multi-file MVPs) and defaults max_tokens to 16384. Every API response is a single OpenAI-shape chat completion — the interactive multi-file Artifact Build Loop (planner + per-file executor + workspace preview) is currently exposed at /codemaster only. Public-API build-loop support ships next; it will bill each planner and per-file call at ₹0.50 (before tier multiplier) alongside the base completion.

• stop — model reached a natural stop or a stop sequence.
• length — hit max_tokens. Increase max_tokens and retry.
• tool_calls — model emitted a tool call (only when agent_mode is on and tools are attached).
• content_filter — output was blocked by safety filters.

Every response carries an X-Request-Id header. Include it in support requests to trace a specific call. Legacy /v1/neurocortex/chat additionally returns a Deprecation header — migrate to /v1/chat/completions.

• Hobby — 2× Lab price, ₹5,000/day cap (default).
• Builder — 1.5×, ₹50,000/day cap.
• Scale — 1.2×, ₹2,50,000/day cap.
• Sovereign — 1.0× (no markup), ₹10,00,000/day cap.

Wallet balance is checked before the upstream call. Low balance returns 402 instantly — no time wasted on a doomed model round-trip. Failed handlers auto-refund. Hard daily caps are atomic at the database.

{ "error": { "code": "insufficient_credits", "message": "Wallet has insufficient credits" },
  "engine": "neurocortex" }

Set webhook_url on a key (Builder+ tier) to receive usage.debit and 402.insufficient_credits events. Payloads are HMAC-signed; verify the X-NeuroCortex-Signature header.