state

Your brain's current allostatic state — rest, engaged, or overloaded — with the load/variance/reserve numbers behind the verdict. A memory product that knows when it's tired.

See Allostatic State for how load is computed and why this is a tool the agent itself should call.

Signature

state() → {
  state: 'rest' | 'engaged' | 'overloaded',
  load: number,
  variance: number,
  reserve: number,
  active_minutes: number,
  window_minutes: number,
  last_active: string | null
}

No parameters.

Example

curl https://mneva.dev/v1/state -H "x-mneva-key: $MNEVA_KEY"

Fresh tenant with no activity:

{
  "state": "rest",
  "load": 0,
  "variance": 0,
  "reserve": 1,
  "active_minutes": 0,
  "window_minutes": 60,
  "last_active": null
}

After a burst of 20 tool calls in one minute:

{
  "state": "overloaded",
  "load": 20,
  "variance": 0,
  "reserve": 0.67,
  "active_minutes": 1,
  "window_minutes": 60,
  "last_active": "2026-05-23T00:14"
}

When to call it

• At the start of a session — see if the brain is fresh or carrying load from a recent run
• Before kicking off a long pipeline — overloaded is the signal to defer, not pile on
• Periodically during heavy work — variance climbs when work is spiky; that's a cue to batch differently

Calling state does NOT contribute to load. The activity tick that drives the verdict is suppressed specifically for /v1/state, so polling it doesn't inflate the very number it returns.

The verdict bands:

• load > 15 OR (load > 4 AND variance > load × 1.5) → overloaded
• load >= 2 → engaged
• otherwise → rest

load is mean tool-calls per minute across active minutes only — a quiet brain reads as rest, not as low-load engaged.

See also

• Allostatic State — concept
• GET /v1/state — REST endpoint