AI Orchestration · Free

Walk away. It gets stronger every cycle.

Drop in a self-running agent loop that does real work, scores itself on a grounded 0–1 reward, and rewrites its own next prompt to get measurably stronger on every ~2-minute cycle — no babysitting.

Download Free

1. See it

fusion · routing

GLM

Cursor

Kimi

DeepSeek

JUDGE

One fused answer

≥ your best single model

2. How it works

Route the task

Models draft in parallel

Cross-verify & fact-check

Strongest-model judge elevates

Deliver one fused answer

3. Why it wins

This product

One model alone

Models working

Several, cross-checking each other

One — a single point of failure

Hallucinations

Caught by cross-verify + a judge

Slip through unchecked

Marginal cost

$0 on memberships you already pay

Per-token API billing

Quality floor

≥ your best single model, by design

Whatever that one model gives

4. Use it to…

Unattended coding marathons

Let the loop attempt, verify, and iterate on coding tasks across many cycles while you're away, returning measurably stronger attempts.

Self-rewriting prompt loops

Let the agent use its 0–1 reward to rewrite its own next prompt each cycle, surfacing which prompt strategies actually score higher.

Verified research-and-report runs

Set it to gather, verify, and score findings on a grounded 0–1 scale every ~2-min cycle, accumulating real results with no hand-holding.

5. The shift it creates

Without it

One model = one point of failure
Hallucinations slip through unchecked
No quality floor you can trust

→

With it

Runs unattended — it schedules its own wake-ups and keeps cycling while you work on something else.
Improves itself measurably
Drop-in and free

6. What's inside

Self-paces — auto-cycles every ~2 min via ScheduleWakeup

Rewrites its own next prompt to get stronger each loop

Grounded 0–1 reward aimed at real outcomes

Runnable cycle.mjs + cycle-rl.mjs (zero-dependency Node)

5-point evolution contract + honesty laws

Instant download — no email, no payment

7. The details

Format

Runnable scripts + docs

Delivery

Instant — no email

License

Yours forever · use on every project

Works with

Any LLM / your own stack

How it actually works

Exactly what it does — and when.

1For a developer running Claude Code who wants continuous autonomous progress → it self-schedules wake-ups every ~2 minutes via ScheduleWakeup, executes one verified work unit, then sleeps until the next cycle.

2For prompt optimization across iterations → cycle-rl.mjs scores each cycle's output on a grounded 0–1 reward scale, feeds that signal back, and rewrites the agent's own next prompt to target weaknesses from the prior loop.

3For teams wanting safety rails on a self-modifying agent → a 5-point evolution contract plus honesty laws constrain what the agent may rewrite about itself, preventing reward-hacking or goal drift.

4For immediate deployment in any Node environment → drop cycle.mjs into your project (zero npm dependencies) and run it under any harness that supports ScheduleWakeup — no email, no payment, instant download.

In practice · Overnight autonomous bug-fixing loop

A solo developer drops cycle.mjs into a Claude Code session at 11 PM, pointing it at a backlog of failing Jest tests. Every ~2 minutes the agent wakes, attempts a fix, runs the test suite, scores 0.0–1.0 based on tests passed, and tightens its own prompt to prioritize the remaining failures. By morning, 14 of 22 tests pass and the evolved prompt reads noticeably different from the original — now biased toward null-checking database returns first.

Questions, answered

✓ Free · no email✓ Yours to own✓ Use on every project✓ Built to ship results