The first nervous system for AI.

Every AI reads text. None reads the human behind it in real time. Nefesh changes that. Any device that reads the body works with Nefesh. Chest strap, smartwatch, webcam, glucose monitor, or your own hardware. Live signal fusion, one API, and your AI adapts to the human instantly. Built for telehealth, coaching, and adaptive learning.

Or try the live demo. No signup required.

Heart Rate
72 BPM
rPPG Camera
Blood flow
Vocal
Pitch stable
Expression
Neutral
Electrodermal (planned)
2.4 μS
Respiratory (planned)
14 br/min
Movement (planned)
Sedentary
Neural / EEG (planned)
Alpha dom.
Sleep (planned)
REM 22%
Metabolic (planned)
SpO2 98%
nefesh
/v1/ingest
signals fused
UNIFIED STATE
28STRESS
relaxed
conf: 0.92
AI ADAPTS
AI

That's worth unpacking. Let's look at what's on your plate — listing things out often reveals it's more manageable than it feels. What are the top three things?

{ "state": "relaxed", "score": 28 }
<48ms·encrypted·edge only
<48ms latency/35+ signals/11 LLM providers/6 MCP tools/GDPR compliant
Works withClaudeCursorVS CodeWindsurfKiro+ 15 more
GDPR compliant
Edge-only processing
Not a medical device
sub-48ms latency

See the difference.

Same user message. Different physiological state. Different AI response.

nefesh live context
v1.2.0

Same input  ·  Different physiology  ·  Different AI

CALM
HRV 72ms
29
ACUTE STRESS
HRV 16ms
83

Integration

Three ways to integrate.

From zero-code gateway to full API control. Pick the integration that fits your stack.

Cognitive Compute Router

Change one URL. Your LLM adapts automatically.

  • 3 integration modes (OpenAI, Anthropic native, Unified)
  • 11 LLM providers supported
  • Zero code changes required
Learn more →

MCP + A2A

Your AI agent reads human state natively.

  • 6 MCP tools for Claude, Cursor, VS Code + 15 more
  • 4 A2A skills for agent-to-agent collaboration
  • Self-provisioning: no signup needed
Learn more →

Direct API + CLI

Full control. One endpoint for every signal.

  • 35+ signal types from any device or sensor
  • Device Registry: register once, never manage sessions
  • CLI: npm install -g @nefesh/cli
Get started →

Why Nefesh.

Not another health API. A fusion engine built on signal science.

Change one URL. Your LLM adapts to the human.

Point your LLM base URL to gateway.nefesh.ai. The gateway reads biometric state, adjusts the system prompt, and forwards to your provider. Three modes: OpenAI-compatible, Anthropic native, and Unified Anthropic for any backend. 11 LLM providers supported. Zero code changes on the app side.

# Before
base_url = "https://api.openai.com"
# After
base_url = "https://gateway.nefesh.ai"
→ Same code. Biometrically-aware responses.

One call. Every signal. One state.

Heart rate alone is noisy. Voice alone lies. Expression alone is ambiguous. Nefesh fuses them all: cardiovascular, vocal, visual, and textual signals. One unified state per API call. No other service does this.

heart_rate: 720.00
+ tone: "tense"0.00
+ expression: "tense"0.00
+ urgency: "high"0.00
→ state: "stressed" · confidence: 0.94

Register once. Never manage sessions again.

Bind a wearable to a user with one API call. After that, send data with just the device ID. The API resolves the user automatically. No session IDs, no manual mapping, no state management on your side.

Step 1: Register (once)
POST /v1/devices
{ device_name: "Polar H10", subject_id: "usr_tom" }
Step 2: Ingest (continuous)
POST /v1/ingest
{ device_id: "dev_abc", heart_rate: 92 }
→ API resolves: dev_abc → usr_tom → correct state

Your AI learns what actually works.

On every call after the first, Nefesh tells the AI whether its previous approach actually reduced stress. Did simplifying the response help? Should it try something different? No other system provides this feedback loop.

Call 172suggested_action: "simplify"
Call 268adaptation_effectiveness: { delta: -4, effective: true }
Call 345adaptation_effectiveness: { delta: -23, effective: true }
→ AI self-improves across the conversation

Your AI remembers what stresses the user.

Nefesh tracks which conversation topics cause stress across turns and sessions. When a resolved trigger reappears, the AI knows to explore deeper. When an active trigger surfaces, the AI knows to tread carefully.

Turn 1"work deadline"72
Turn 3"work deadline"68
Turn 5"work deadline"31
Turn 8"relationship"81
AI prompt:
ACTIVE: relationship — extra care.
RESOLVED: work deadline — can explore deeper.

Solutions

Built for regulated industries.

Nefesh is used by development teams building telehealth platforms, AI coaching tools, workplace wellness programs, and adaptive learning systems. If your product interacts with humans under stress, Nefesh gives your AI the context to respond appropriately.