Optimizer Documentation - PromptAsCode

Overview

The Optimizer compresses your prompts to use fewer tokens while preserving their semantic meaning and effectiveness. Reduce API costs and fit more context into limited token windows without sacrificing prompt quality.

Token Reduction

20-40% typical savings

Cost Savings

Immediate ROI

90%+ Meaning

Semantic preservation

How to Use

1
Paste Your Prompt - Enter the prompt you want to optimize into the editor.
2
Select Mode - Choose Safe Mode (conservative) or Aggressive Mode (maximum compression).
3
Click Optimize - The optimizer analyzes and compresses your prompt.
4
Review Changes - See the optimized version alongside token savings and similarity score.
5
Compare & Test - Use Prompt Diff to see changes, test in Playground to verify quality.
6
Apply or Iterate - Use the optimized version or adjust settings and try again.

Optimization Modes

Safe Mode (Conservative)

Preserves 95%+ semantic similarity. Focuses on removing redundancy and obvious inefficiencies without changing meaning.

Typical Savings

15-25% tokens

Similarity

95-99%

Aggressive Mode (Maximum)

Targets 90%+ semantic similarity. Uses advanced compression including sentence restructuring and abbreviation where appropriate.

Typical Savings

30-45% tokens

Similarity

90-95%

Which Mode to Choose

Safe Mode: Production prompts, critical instructions, safety guidelines
Aggressive Mode: Long examples, verbose descriptions, context documents

Optimization Techniques

What the Optimizer Does

Remove redundancy: Eliminate repeated instructions and information
Simplify phrasing: "In order to" → "To", "Make sure to" → "Must"
Consolidate lists: Merge related bullet points
Tighten examples: Shorten while preserving pattern
Remove filler: Cut words that don't add meaning
Restructure sentences: More efficient sentence construction

Example Optimization

Before (147 tokens):
You are an AI assistant that helps users with their
questions. You should always be helpful and provide
accurate information. When answering questions, make
sure to be clear and concise in your responses. If
you don't know the answer to something, it's okay
to say that you don't know.

After (89 tokens, 40% reduction):
You are a helpful AI assistant. Provide accurate,
clear, concise answers. Say "I don't know" when
uncertain.

Semantic Similarity: 94%

Quality Preservation

The optimizer measures semantic similarity to ensure quality:

95-100%

Excellent. Safe to use as direct replacement.

90-95%

Good. Review and test before production.

<90%

Caution. Significant changes, test thoroughly.

Always Verify

While semantic similarity is a good guide, always test optimized prompts in the Playground before using in production. Some nuances may not be captured by similarity scores.

AI Expert Use Cases

Cost Reduction at Scale

For high-volume applications, even 20% token reduction translates to significant cost savings. A prompt used 1M times/month with 100 tokens saved = 100M fewer tokens.

Context Window Management

When prompts approach context limits, use the Optimizer to create headroom. Essential for RAG applications where you need space for retrieved documents.

Latency Optimization

Fewer tokens = faster responses. Optimize prompts for latency-sensitive applications like chatbots and real-time systems.

Prompt Library Maintenance

Periodically run your prompt library through the Optimizer to identify accumulated verbosity and optimize for efficiency.

Tips & Best Practices

Pro Tips

Start with Safe Mode, only use Aggressive if needed
Always compare with Prompt Diff before applying
Test optimized prompts in Playground
Benchmark before and after to verify quality
Keep the original version in version control
Don't optimize prompts that are already minimal

What NOT to Optimize

Be careful optimizing:

Safety and guardrail instructions (precision matters)
Complex few-shot examples (details may be important)
Legal or compliance text (exact wording may be required)
Prompts already under 100 tokens (diminishing returns)