Real-time speech-to-speech, voice-to-text, and text chat orchestration. Failover, RAG, structured outputs, cost tracking — one API, zero infrastructure.
<div id="chat"></div> <script src="https://api.neuralspeak.ai/sdk/neuralspeak.js"></script> <script> NeuralSpeak.init({ apiKey: 'ns_pub_...', containerId: 'chat' }); </script> // That's it. All config lives in the dashboard.
Same SDK. Same dashboard. Same billing. Same API keys. Pick the mode per service.
Full speech-to-speech conversations. User speaks, AI responds with natural voice. Real-time NDJSON streaming with parallel TTS and sub-2s perceived latency.
LLM-powered chatbot with RAG knowledge base. Markdown rendering, structured outputs, zero audio overhead. Deploy customer support, sales, or FAQ bots instantly.
User speaks, AI responds with text. Perfect for medical intake forms, dictation, accessibility tools, and voice-driven search.
You send audio or text. The engine handles the rest — routing, failover, caching, analysis — and streams back the response.
Start simple and go deeper when you need to. Both paths lead to production.
No tech skills needed. Set it up yourself in minutes — the dashboard guides you step by step. Add 3 lines of code to your product, and you're live.
Pick a mode (voice/text/both), write prompts, select providers, upload docs, customize colors. All visual, all clicks.
One <script> tag + one line of JavaScript. The SDK auto-loads all settings from your dashboard.
Voice agent or chatbot on your site. Change prompts, providers, or theme in dashboard — instant update, zero redeploy.
For teams who need dynamic context, custom UIs, user-specific data, and deep integration.
Your server holds the secret key (ns_sec_). Proxy requests to NeuralSpeak API. Inject CRM data, user profiles, order history mid-conversation.
Build your own UI with the headless REST API. Or use the SDK with custom themes, layouts, and callbacks. Structured outputs trigger your automations.
Failover chains, rate limiting, webhooks, usage analytics, session history, per-key CORS — production-grade from day one.
STT, LLM, and TTS each have failover chains. Groq down? Deepgram takes over. Azure timeout? ElevenLabs kicks in. Zero downtime for your users.
Multiple sentences synthesized simultaneously via Semaphore(3), reordered for sequential playback. Sub-2s perceived latency.
Upload PDFs, TXT, DOCX. Auto-chunked, embedded via OpenAI, stored in pgvector. Top-K retrieval per query. Configurable per service.
Hidden LLM analysis after each exchange. Lead scoring, medical triage, ticket categorization. JSON schema validation, preset templates, webhook delivery.
Bot proposes actions, user confirms, system executes. Whitelist-based, confirmation flow, audit log. Our own help bot showcases this live.
Transparent cost tracking per exchange: STT + LLM + TTS + RAG. 5 subscription tiers, NST top-up packs (never expire), CSV export, projected costs.
15+ settings: themes (dark/light/auto/custom), layouts (full/compact/minimal), positions, colors, branding, custom CSS, word highlight animation. All from dashboard.
Public keys (ns_pub_) for frontend, secret keys (ns_sec_) for backend. Per-key project binding, IP allowlist, domain CORS, TTL expiry, rate limiting, scope enforcement.
Session completed, usage threshold, analysis ready. HMAC-SHA256 signing, delivery logs, retry logic. Push events to your backend in real-time.
Groq Whisper, Deepgram Nova-3, OpenAI GPT-4o, Anthropic Claude, Azure Speech, ElevenLabs. Pick per service. Switch providers with zero code changes.
4 charts (NST, exchanges, cost, prediction), per-service breakdown, session transcripts, request logs with pipeline detail. Filter by period, export CSV.
Inject user-specific data mid-conversation via PUT /sessions/{id}/context. Personalize responses with CRM data, user profiles, order history — in real time.
Same platform, different prompts. Configure once, deploy everywhere.
Voice or text support agent with RAG over your docs. Auto-categorize tickets via structured outputs.
Language tutors, exam prep, interactive lessons. Voice-based with real-time translations and proficiency assessment.
Medical intake forms, symptom collection, triage. Structured outputs for EHR integration. Never diagnoses — just collects.
Lead qualification bots, product demos, pricing queries. Dynamic context from CRM. Intent scoring via structured outputs.
Plan comparison, account info, top-up assistance. Text chatbot with RAG over offer documents. Multi-language support.
Restaurant concierge, hotel booking, menu recommendations. Voice agent in any language with time-aware prompts.
Anywhere a voice or text AI can help — HR onboarding, legal intake, insurance claims, real estate, fitness coaching, and more. Same platform, your prompts.
Four ways to integrate. All battle-tested with real showcase apps.
Static HTML + SDK. All config in dashboard. Zero backend needed.
<script>
NeuralSpeak.init({
apiKey: 'ns_pub_...',
containerId: 'chat',
});
</script>
SDK with custom theme, labels, layout. Override server config locally.
NeuralSpeak.init({
apiKey: 'ns_pub_...',
containerId: 'chat',
theme: { accentColor: '#e11d48' },
labels: { ready: 'Tap to talk' },
});
Your server proxies requests. Inject CRM/user data mid-conversation.
// Your backend (ns_sec_ key):
fetch(API + '/sessions/' + sid
+ '/context', {
method: 'PUT',
body: JSON.stringify({
mode: 'prepend_next',
text: userContext
})
});
No SDK. Build your own UI. Full pipeline control.
curl -X POST \
$API/v1/sessions/message \
-H "Authorization: Bearer $KEY" \
-d '{"text":"Hello",
"project_id":"..."}'
Billed in NST (NeuralSpeak Tokens). Voice exchange ≈ 3 NST. Text exchange ≈ 0.5 NST. No hidden fees.
Need more NST? Buy top-up packs anytime — they never expire. See top-up pricing
From 3-line embed to full backend integration. Comprehensive guides for every skill level.
Get your first AI service running in under 5 minutes.
That's it. Your AI is live. All configuration lives in the dashboard — no code changes needed.
Three auth levels for different use cases.
Public keys are safe to expose in HTML. They can only start conversations. Secret keys have full API access — never put them in frontend code.
30+ endpoints for complete platform control.
/v1/sessions/start Create a new session/v1/sessions/{id}/speech Send audio, get streaming response/v1/sessions/message Text chat (no audio)/v1/sessions/initial Generate AI greeting/v1/sessions/{id}/history Conversation history/v1/sessions/{id}/context Inject context mid-session/v1/auth/register Create account/v1/auth/login Get JWT tokens/v1/projects List services/v1/projects Create service/v1/projects/{id} Update service config/v1/projects/{id}/context Upload knowledge base file/v1/usage Usage analytics/v1/sessions List all sessions/v1/api-keys Generate API key/v1/sdk/config Widget config for SDKFull interactive reference with schemas and try-it available in the dashboard.
Zero-dependency embeddable widget.
SDK auto-fetches config from dashboard via GET /v1/sdk/config. Local options override server defaults.
Real-time response format.
Structured outputs arrive in the complete event's analysis field — invisible to end users.
Real-time event notifications.
HMAC-SHA256 signed. Delivery logs in dashboard. Configure in Webhooks settings.
Hidden AI analysis after each exchange.
Invisible to end users. Perfect for CRM integration, ticket routing, compliance monitoring.
Upload docs, AI answers from your content.
Configurable per service: chunk size (100-2000), overlap (0-500), top-K (1-20), embedding model. Test queries in dashboard.
Connect Claude Desktop to NeuralSpeak via MCP (Model Context Protocol). Create services, configure prompts, upload knowledge base, set up webhooks — all through natural conversation.
pip install neuralspeak-mcp → add config → restart Claude
Free tier included. No credit card required. Deploy your first voice agent or chatbot in under 5 minutes.