Prompt Compiler transforms messy natural language ideas into structured, optimized System Instructions and User Prompts — powered by an LLM engine with local heuristic fallback.
The engine analyzes your intent and produces four output layers:
- System Prompt — Persona, role, constraints, and output format rules for the target AI.
- User Prompt — Structured, clean task definition derived from your input.
- Execution Plan — Step-by-step logic decomposed from your request.
- Expanded Prompt — A single combined prompt ready to paste into any LLM chat.
Switch between the output tabs in the UI to view each layer. A one-click copy button is available on every tab.
The Conservative toggle controls how aggressively the compiler interprets your input.
| State | Behavior |
|---|---|
| ON (default) | Stays strictly grounded in what you wrote. No invented details, no extra requirements, no hallucinated libraries or APIs. Short inputs produce minimal, sensible prompts. Missing information triggers clarifying questions instead of fabricated answers. |
| OFF | Aggressive optimization mode — expands context, infers likely best practices, adds scaffolding. Use when you want the compiler to fill gaps and produce richer prompts. |
The toggle is available in both the web app (top-right header) and the browser extension popup. Its state is persisted locally (localStorage / chrome.storage).
Why this matters: Without conservative mode, a one-word input like "hello" could generate an unrelated Python code snippet. With conservative mode on, the same input produces a clean, friendly greeting prompt.
Describe a role or autonomous task, and the Agent Generator will produce a complete, constraint-driven system prompt for an AI agent.
- Single Agent — Generates a focused, single-role agent prompt with strict boundary conditions.
- Multi-Agent Swarm — Toggle the "Multi-Agent" flag to generate a cooperative swarm architect prompt instead, describing how multiple specialized workers should coordinate.
After generating an agent, an Export section appears below the output. It converts your agent's system prompt into ready-to-run framework code:
| Framework | Output |
|---|---|
| Claude SDK | Python code using the anthropic client |
| LangChain | Python agent with LangChain's ChatPromptTemplate |
| LangGraph | Python graph definition with node/edge structure |
Each framework tab produces both a Python Code file and a YAML Config file. Hover over the code block to reveal the copy button.
Describe a capability in plain English, and the Skill Generator translates it into a structured tool definition.
- Produces a complete Input Schema and Output Schema in valid JSON.
- Generates a stringified skill definition ready for LangChain, OpenAI functions, or custom agent frameworks.
After generating a skill, an Export section appears below the output. It wraps your skill definition in framework-specific code:
| Format | Output |
|---|---|
| LangChain Tool | Python @tool decorated function + JSON schema |
| Claude tool_use | JSON config compatible with the tools parameter in Anthropic's API |
Each format produces both a Python Tool file and a JSON Schema / Config file. Hover over the code block to reveal the copy button.
Compresses your prompt by 20–30% without losing meaning, logic, or variables. Useful when working near context window limits.
A/B test your prompts against compiled versions:
- Raw vs. Compiled — Side-by-side quality comparison.
- Auto-Judge — Real-time scoring of response quality, relevance, and clarity.
- Visual Metrics — Improvement percentages and radar charts.
Upload project files (PDF, MD, TXT, code) to ground the compiler in your domain context.
- Context Manager — Drag-and-drop your brand guidelines, API docs, or any reference material.
- Agent 6 (The Strategist) — Scans uploaded files for relevant facts and injects them into prompt generation.
- Agent 7 (The Critic) — Cross-references the generated prompt against your knowledge base and blocks hallucinated facts.
- Intelligent Caching — Local SQLite vector store (
~/.promptc_index_v2.db) for instant retrieval without re-uploading.
git clone https://github.com/madara88645/Compiler.git
cd Compiler
# Backend
pip install -r requirements.txt
# Frontend
cd web && npm install && cd ..cp .env.example .envEdit .env:
OPENAI_API_KEY=sk-your-actual-key
OPENAI_BASE_URL=https://api.openai.com # optional
# Required for Benchmark Arena:
GROQ_API_KEY=gsk_your_groq_key
# Compiler mode: conservative (default) or default
PROMPT_COMPILER_MODE=conservativeWindows (one-click): double-click start_app.bat
Manual:
# Terminal 1 — Backend
python -m uvicorn api.main:app --reload --port 8080
# Terminal 2 — Frontend
cd web && npm run devOpen http://localhost:3000.
- Type your idea in the input box (any page).
- Click Generate — the LLM analyzes your intent and produces structured output.
- Review the output tabs: System, User, Plan, Expanded.
- Copy the result with the copy button, or use the Export button (Agent / Skill pages) to get framework-ready code.
- Toggle Conservative to control how strictly the compiler stays within your original text.
api/ FastAPI endpoints (compile, agent-generator, skills-generator, optimize…)
app/
compiler.py Chain-of-Responsibility heuristic compiler (V1 + V2)
emitters.py Prompt rendering layer (system, user, plan, expanded)
llm_engine/ HybridCompiler, WorkerClient, LLM prompts
prompts/
worker_v1.md Standard compiler system prompt
worker_conservative.md Conservative (anti-hallucination) system prompt
heuristics/ Local risk detection and offline parsing
rag/ SQLite FTS5 RAG index and context retrieval
web/
app/
page.tsx Main compiler UI
agent-generator/ Agent Generator page + ExportPanel
skills-generator/ Skill Generator page + SkillExportPanel
benchmark/ Benchmark Playground
optimizer/ Token Optimizer
components/ Shared UI components
tests/ Full test suite (200+ tests, all offline-safe)
Copyright © 2026 Mehmet Özel. All rights reserved.
Licensed under the Apache License 2.0.
For managed/hosted service inquiries: mehmet.ozel2701@gmail.com
Self-hosting is free and always will be.
Built with ❤️ for Prompt Engineers.