[nevrai]
· 6 min read

nevrai: LEGO for AI Products

What if adding AI to your app was as simple as npm install?

Not “spend two weeks building LLM routing, error handling, auth, and billing.” Actually simple — one API call and you have a working LLM chat with automatic failover across three providers.

That’s what nevrai does.

The Problem Every AI Product Rebuilds

You have an idea for an AI product. Maybe a customer support bot, a research tool, or an internal knowledge assistant. You start building, and a week in you’re deep in infrastructure code that has nothing to do with your actual product:

  1. LLM routing — which model to use? What happens when it’s unavailable? How do you switch between providers without rewriting everything?
  2. Error handling — rate limits, timeouts, malformed responses, provider outages at 3am.
  3. User frustration detection — your users are getting angry, and the bot keeps responding with the same cheerful tone.
  4. Structured data extraction — the LLM gives you prose when you need JSON. Every time.
  5. Document creation — turning AI output into something shareable and verifiable.

Every AI startup solves these problems from scratch. Most solve them poorly. Some never ship because they get stuck on infrastructure.

5 Modules, 5 API Calls

nevrai is a set of independent API modules. Use one, use all five — they work together but don’t depend on each other.

1. LLM Runtime

Drop-in chat API with a 3-tier cascade: Groq (fast, free) → OpenRouter (broad model selection) → bootstrap fallback (always available). You send messages, nevrai picks the fastest available model and streams the response via SSE.

curl -X POST https://api.nevrai.com/v1/chat \
  -H "Authorization: Bearer nvr_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [
      {"role": "user", "content": "What is JTBD?"}
    ]
  }'

Two routing strategies out of the box: speed_first (default, picks the model with lowest latency) and quality_first (picks the most capable model within budget). You don’t manage API keys for Groq, OpenRouter, or other providers. nevrai handles all of that.

2. Escalation Detector

Detects when a user is frustrated, stuck, or about to leave. Uses a two-tier approach: fast regex patterns catch obvious signals (“this is useless”, “nothing works”), then LLM classifies ambiguous cases. Returns a structured verdict with trigger type and recommended action.

curl -X POST https://api.nevrai.com/v1/escalation \
  -H "Authorization: Bearer nvr_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "message": "This is useless, nothing works!",
    "history": []
  }'

Five trigger types: explicit request, negative tone, tone mismatch, stagnation (user going in circles), and repetition. Works in English and Russian. Typical use: when escalation is detected, route the user to a more capable — and more expensive — model.

3. Data Extractor

Send raw text — a chat conversation, interview transcript, support ticket — get structured JSON with up to 19 JTBD data points: pains, jobs, triggers, audience segments, willingness to pay, and more.

curl -X POST https://api.nevrai.com/v1/extract \
  -H "Authorization: Bearer nvr_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "text": "User: I hate waiting 30 min for a charger...",
    "language": "en"
  }'

This is the module that turns conversations into data. Instead of reading 500 support tickets manually, run them through the extractor and get a structured dataset of what your users actually need.

4. PDF Publisher

Send markdown, get back a signed PDF with a unique document ID. The document is verifiable — anyone can confirm its authenticity in the nevrai registry.

curl -X POST https://api.nevrai.com/v1/pdf \
  -H "Authorization: Bearer nvr_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "content": "# Report\n\nKey findings...",
    "format": "markdown"
  }'

Use case: your AI generates a report, analysis, or recommendation. You need to share it with stakeholders who want a real document, not a chat screenshot. PDF Publisher handles rendering, signing, and hosting.

5. Discovery Engine

Multi-round market research via API. Give it a query and it searches the web and Telegram channels, filters results through an LLM, and returns structured insights. Results are cached for 90 days — repeat queries for the same niche return instantly.

curl -X POST https://api.nevrai.com/v1/research \
  -H "Authorization: Bearer nvr_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "AI text detection market",
    "depth": "quick"
  }'

The heaviest module — runs asynchronously and can take 30-60 seconds for deep search. The quick depth returns results in under 10 seconds with fewer sources.

Why the Cascade Matters

The LLM Runtime cascade is the foundation of nevrai’s reliability story. Here’s what happens on every request:

  1. Groq — fastest inference available. Time to first token under one second. Free models. If Groq is up, your users get responses in under a second.
  2. OpenRouter — if Groq is unavailable or rate-limited, nevrai automatically falls back to OpenRouter, which aggregates dozens of model providers. Same API call, same response format.
  3. Bootstrap fallback — if both Groq and OpenRouter are down (it happens), nevrai falls back to a known-good model configuration that’s always available.

Dead models are automatically blacklisted for one hour after a 403, 404, or empty response. Model availability updates every 12 hours from provider APIs. You don’t manage any of this.

Result: your application gets 99.9%+ effective uptime for LLM calls, even when individual providers go down.

Pricing

PlanPriceRequests/moModules
Free$01,000Chat + Escalation
Starter$49/mo10,000All 5 modules
Pro$149/mo100,000All 5 modules
Business$399/mo1,000,000All 5 + SLA + dedicated support

The free tier isn’t a trial. No expiration. If 1,000 requests/month covers your use case, it’s free forever.

Getting Started

  1. Go to nevrai.com/dashboard
  2. Sign in with Google
  3. Create an API key
  4. Make your first call

No credit card required for the free tier. No SDK required — it’s just HTTP.

curl -X POST https://api.nevrai.com/v1/chat \
  -H "Authorization: Bearer nvr_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"messages":[{"role":"user","content":"Hello"}]}'

If you’re building an AI product and spending more time on infrastructure than on your actual product — nevrai exists so you can stop doing that.

Get your API key