FOR AI BUILDERS

Your agent knows their calendar. It doesn't know their mind.

AI agents schedule meetings, triage tasks, and draft replies, blind to whether their human is cognitively ready for any of it. WhenPeak is the performance layer for the agentic economy. One API call gives your agent the context it's missing.

< 1s stateless response, fits inside an agent loop1 endpoint, OpenAPI spec, any LLM
Agent loop
01 · User wakes
02 · Agent calls WhenPeak
03 · Receives peak / dip / DPS
04 · Schedules the day

Agents optimise for time. They ignore cognitive state.

The problem

Every scheduling agent makes the same hidden assumption: that all hours are cognitively equal. Book the strategy session at 2pm? Fine. Schedule three back-to-back negotiations after a bad night's sleep? No objection. Send the "reply to this important email" nudge during the post-lunch crash? Sure.

The agent has no model of the human's capacity. It treats a calendar slot as a calendar slot. But a 10am deep-work block and a 2pm deep-work block produce measurably different output, research shows working memory and processing speed vary 20-40% across the day.

The fix

WhenPeak gives your agent a 24-hour model of its user's cognitive capacity, derived from sleep data and the Two-Process Model of sleep/wake regulation (Borbély 1982).

With one API call, your agent knows:

  • · The user's peak analytical window (when to schedule deep work)
  • · Their natural dip (when to route admin and email)
  • · Their secondary peak (when to schedule calls and collaboration)
  • · Their daily readiness score (when to suggest rest instead of pushing)

The agent stops scheduling blind.

Six things your agent can do with one prediction

Schedule deep work at peak

Route the user's most cognitively demanding tasks into their predicted peak window. The same task, done at peak vs dip, can take 15-30% less time with 20-40% fewer errors.

📬

Protect the dip

Automatically schedule low-cognitive overhead: email, admin, routine reviews, into the predicted post-lunch dip, instead of wasting peak hours on it.

🛡

Avoid high-stakes scheduling on low-readiness days

When the DPS is low, the agent can flag or reschedule high-stakes meetings: negotiations, board presentations, interviews, where fatigue-driven errors are most costly.

😴

Suggest rest proactively

A low DPS isn't just a scheduling input, it's a wellbeing signal. The agent can proactively recommend lighter days, protecting the user from burnout cycles.

🎯

Prepare for the moments that matter

Given the user's peak window and an upcoming high-stakes event, the agent can time preparation, reminders, and even caffeine suggestions to land the user at peak when it counts.

🔔

Personalise its own timing

The agent can time its own notifications and check-ins to match the user's cognitive state, engaging during receptive windows, staying quiet during deep focus.

Integrate in under an hour

Path 1

Direct API (any framework)

Request
curl -X POST https://api.whenpeak.com/api/v1/predict \
  -H "Content-Type: application/json" \
  -d '{"sleep_time":"23:00","wake_time":"07:00","sleep_quality":"good"}'
Response
{
  "dps": 74,
  "peak_1": { "time": "10:00" },
  "dip":    { "time": "14:00" },
  "peak_2": { "time": "17:00" },
  "chronotype": "Third Bird",
  "confidence": "low"
}
Path 2

LLM function calling

Pass the OpenAPI spec directly to your model's tool definitions.

https://api.whenpeak.com/openapi.json

Works with OpenAI function calling, Anthropic tool use, LangChain, and any framework that accepts OpenAPI tool schemas.

Path 3

Native integrations

ChatGPT GPT ActionLive

Reference implementation for GPT builders.

Claude MCP skillLive

Remote MCP server deployed and callable.

Built for apps with real users

Building a VA, a productivity app, or an agent platform with many users? Provision isolated end-user accounts under your developer key. Each user gets their own sleep history, chronotype, and personalised curve that improves over 30 days.

Architecture
Your developer key (pk_live_...)
  ├── End user A → own sleep history, own curve
  ├── End user B → own sleep history, own curve
  └── End user C → own sleep history, own curve
        ↓
  All consuming from your tier's monthly quota
1,600 DAUs

supported on Growth tier (50K calls/mo at 1 prediction/user/day)

16,000 DAUs

supported on Scale tier (500K calls/mo)

30 days

to full personalisation per end-user

Why this works: the Two-Process Model

WhenPeak isn't a heuristic. It implements the Two-Process Model of sleep/wake regulation, the dominant framework in chronobiology since Borbély (1982) and empirically validated by Dijk & Czeisler (1995).

Process C models the circadian alerting signal, phased to the user's chronotype. Process S models homeostatic sleep pressure building from wake time. Their interaction produces the characteristic daily curve: a morning analytical peak, a post-lunch dip, and a late-afternoon secondary peak. The model is what lets WhenPeak return a defensible prediction rather than a guess.

Research basis
  • Borbély (1982)
    "A two process model of sleep regulation"
    Human Neurobiology, foundational framework
  • Dijk & Czeisler (1995)
    "Contribution of the circadian pacemaker and the sleep homeostat to sleep propensity"
    Journal of Neuroscience, empirical validation
  • 222 unit tests
    "Algorithm validated against expected curve shapes"
    Across chronotypes and sleep conditions
Live demo

See what your agent would receive

< 1s

Stateless prediction latency, fits in an agent loop

1 call

To add cognitive awareness to any agent

20-40%

Variance in cognitive capacity across the day the agent now sees

2

Live native integrations (ChatGPT, Claude) + open OpenAPI

Common questions

Call the WhenPeak API endpoint POST /api/v1/predict with the user's sleep time, wake time, and quality. You receive their cognitive peak window, natural dip, secondary peak, and a Daily Performance Score (DPS). The public endpoint requires no authentication for stateless predictions; authenticated endpoints provide personalised predictions that improve over 30 days. The OpenAPI spec is published for direct function-calling integration with any LLM.

Give your agent a cognitive layer.