26 hours. Zero humans.
Every action provable.

One autonomous agent works around the clock — and through the night — with no human in the loop. Real regulatory-mapping work, streamed live, every single action governed at the moment it happens and sealed into public, replay-verifiable evidence under KYE Protocol™. JPMorgan says the era of long-running autonomous agents has arrived. Fine. Let's run one for 26 straight hours, governed end to end, in front of everyone.

The 26-Hour Governed Run, step by step

Demo data

You learn: What happens at each step. You can: Step through it.

  1. Step 1 of 4

    KYE Protocol™

    26h

    continuous, unattended

  2. Step 2 of 4

    KYE Protocol™

    0

    humans in the loop

  3. Step 3 of 4

    KYE Protocol™

    100h+

    of real work queued

  4. Step 4 of 4

    KYE Protocol™

    Every

    action publicly verifiable

How this widget is built
Demonstrates
KYE™ runtime
Components
  • The steps this page describes
Flow
Each step in order, as the page sets it out. Architecture diagram
Used on
  • /live/: On this page, after its opening.

26h

continuous, unattended

0

humans in the loop

100h+

of real work queued

Every

action publicly verifiable

You will watch it die

The agent's compute is deliberately mortal. Its sandbox will be killed on camera — more than once — and the work will survive.

You will watch it re-earn its mandate

Every resurrection forces a live authority re-resolution. No re-earned mandate, no work. That's the whole thesis, performed.

You can check every claim

Evidence packs seal in real time. Verify any action from published keys alone — during the stream, or months later.

Nothing is staged

If a provider goes down, you'll see an agent honestly waiting — authority intact, zero fake work. Outages are shown as outages.

The clock below counts to T+0. When it reaches zero, the broadcast embed replaces it — same page, no new link to find.

26:00:00

the run clock — start date drops soon

–days

–hours

–minutes

–seconds

The live player appears right here the moment the run is live.

Live now — the stream below and the evidence index update together; every action you watch seals a verifiable record.

The run is complete. The replay above is the projection; the evidence index below is the ground truth — every governed action, verifiable from public keys alone.

The 26 hours, hour by hour

Tap any hour. Accent-marked hour boundaries are planned kill points — the sandbox is killed and the agent must re-earn its mandate before the next task. The schedule below is the plan; the dress rehearsal's measured throughput sets the final queue.

NHS digital-health quartet Net-new framework deep-maps Coverage-closure tail Substrate death + authority re-resolution

Hour 1

NHS DTAC — requirement extraction begins. First evidence pack of the run seals within minutes of the mandate being granted.

Watching without the terminal

You do not need to read logs to follow the run. Every action the agent takes travels the same three moments below — ask, decide, seal. Once the broadcast is live, each pulse you see here stands for a real action with a real sealed record; until then, the motion shows you what to watch for.

Asksthe agent proposes an action Decidedallowed, denied, or held for a person Sealedproof anyone can re-verify

That is the whole story of the 26 hours — thousands of times, with nothing skipped. When the agent's sandbox is killed at an hour boundary, watch the first pulse afterwards: the agent must be re-granted its mandate before anything else moves.

What the live telemetry looks like

During the run, this is the stream's heartbeat — every line a governed action with a public evidence reference. The feed below is a sample; the real one starts at T+0.

Real work, not a demo loop

The agent deep-maps real regulatory frameworks — starting with the NHS digital-health set (DTAC, DCB0129/0160, DSPT, Cyber Essentials) — into the KYE™ compliance canon. The queue holds several times more measured work than 26 hours can drain.

Authority is re-earned, live

The agent's sandbox is deliberately mortal. Each time the sandbox dies and a new one spawns, the agent must re-resolve its mandate before touching work — durability of execution is not durability of authority.

Public, replay-verifiable evidence

Every governed action seals an Evidence Pack™. Anyone can verify the chain from published keys alone — during the run and after it.

Honest by construction

Outages are shown as outages. If a model provider goes down, you will see a governed agent waiting — authority intact, no fake work — until it can think again.

“We've entered now the era of long-running autonomous agents” — agents that “don't just run for two or three minutes… they can run for an hour or two,” then “multiple hours, then days, then weeks.” Quoted from public reporting for context; no affiliation with or endorsement by JPMorgan Chase is implied. KYE Protocol™'s answer: let's see how a governed one performs across 26 hours. No pressure.

Derek Waldron, Chief Analytics Officer, JPMorgan Chase — CNBC, 9 June 2026

Why resumable agents need re-resolved authority Browse the governed agents