RIDER — Autonomous Prompt Optimization

Python client for RIDER (Reflective Iterative Diversity-Enhanced Reasoning) — a metaheuristic prompt optimization engine that evolves prompts through evolutionary algorithms, multi-armed bandits, and error-directed refinement.

17 WIN / 0 LOSE across 6 benchmarks (GSM8K, AG_News, SQuAD 2.0, CommonGen, XSum, CodeSearchNet) on 3 LLM models (Claude Haiku 4.5, GPT-4.1, DeepSeek V3.2). Published at FSE 2026 (CORE A*) and FRUCT 39 (IEEE).

See it in action

🔴 Before — 3 words, vague, unusable

Write a summary

🟢 After RIDER — 185 words, structured, production-ready

ROLE: Expert technical writer specializing in distilling complex content
EXPERIENCE: 10+ years in information architecture and content strategy

TASK: Create a comprehensive yet concise summary of the provided content

METHODOLOGY:
  1. Identify the core thesis and 3-5 key supporting arguments
  2. Extract critical data points, statistics, and evidence
  3. Prioritize information by actionability and impact
  4. Structure using inverted pyramid (most important first)
  5. Verify no critical information is lost in compression

FORMAT: 3-5 sentences. Lead with the conclusion.
TONE: Neutral, factual, no editorializing

CONSTRAINTS:
  - No filler phrases ("In this article...", "It is worth noting...")
  - No personal opinions or subjective assessments
  - Preserve all numerical data and proper nouns exactly
  - If the source contradicts itself, note the contradiction

ANTI-PATTERNS:
  - Do NOT start with "This article discusses..."
  - Do NOT use bullet points (prose only)
  - Do NOT add information not present in the source

Result: 185 words · +6000% expansion · +167% quality improvement

fix the bug

ROLE: Senior software debugging specialist with 15+ years of experience
      in systematic root cause analysis.

TASK: Diagnose and fix the reported bug using a structured methodology.

METHODOLOGY:
  1. REPRODUCE:  Identify exact steps to trigger the bug; document
                 expected vs. actual behavior.
  2. DIAGNOSE:   Binary-search through recent changes; check logs,
                 stack traces, and recent commits.
  3. FIX:        Implement the minimal targeted fix; add a regression
                 test that would have caught it.
  4. VERIFY:     Run the full test suite and check adjacent edge cases
                 before closing the issue.

CONSTRAINTS:
  - Do not silence errors with try/except — address the root cause.
  - Do not add defensive code for scenarios that cannot happen.
  - The fix must not break any existing tests.

Result: 120 words · +3900% expansion

Try it yourself: riderprompt.com has a live demo that runs RIDER on your prompt in under 30 seconds.

Installation

pip install riderprompt

Quick Start

Get your API key at riderprompt.com, then:

from rider import Rider

client = Rider(api_key="rider_sk_...")

result = client.optimize(
    prompt="Write a summary of this article",
    mode="standard",
)

print(result.optimized_prompt)
print(f"Improvement: {result.improvement:.1f}%")
print(f"Fitness score: {result.fitness:.2f}")
print(f"Duration: {result.duration_seconds:.1f}s")

Optimization Modes

Features

• 🧬 Evolutionary algorithms — 9 operators (EDA mutation, differential evolution, genetic crossover, semantic paraphrase, etc.)
• 🎰 Adaptive operator selection — Windowed Thompson Sampling with Bayesian elimination
• 🎯 Error-directed evolution — feeds misclassified examples back into generation
• ⚡ Evaluate-first (μ+λ) retention — diversity tiebreaking
• 🔄 8 exclusive protocols — PRISM, NEXUS, PHASE REACTOR, OPERATOR FORGE, ECHO, CHIMERA, VORTEX, MARS CONFIG

Examples

Optimize a prompt with custom validation data

result = client.optimize(
    prompt="Classify this text as positive or negative",
    mode="ultra",
    validation_examples=[
        {"input": "I love this!", "expected": "positive"},
        {"input": "This is terrible", "expected": "negative"},
    ],
)

Async usage with multiple prompts

import asyncio
from rider import AsyncRider

async def optimize_many(prompts):
    client = AsyncRider(api_key="rider_sk_...")
    tasks = [client.optimize(p, mode="light") for p in prompts]
    results = await asyncio.gather(*tasks)
    return results

Pricing

• Free — 3 optimizations/day (Light + Blitz modes)
• Pro ($9.99/mo) — 50/day, all 4 modes, REST API
• Ultra ($29.99/mo) — Unlimited, priority queue

See pricing

Citation

If you use RIDER in your research, please cite:

@inproceedings{dragomirov2026rider,
  title={{RIDER}: Evolutionary Prompt Optimization with Adaptive Operator Selection for Software Engineering},
  author={Dragomirov, Daglar and Kulin, Nikita and Muravyov, Sergey and Makarov, Ilya and Sukhorukov, Daniil and Mozikov, Mikhail},
  booktitle={Companion Proceedings of the 34th ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering (FSE Companion '26)},
  year={2026},
  doi={10.1145/3803437.3807393},
}

Links

• Website: riderprompt.com
• API Documentation: riderprompt.com/api-docs
• Developer Portal: riderprompt.com/developer
• Paper: FSE 2026 Companion Proceedings

License

MIT License — see LICENSE file.