macOS menu bar

Before your words
reach the model,
the work is done.

Seven specialized stages run invisibly between you and any AI. You type a rough idea. PromptCraft delivers a precision-engineered request. No context writing. No repeated scaffolding. No wasted tokens.

Download for Mac — Free Pro · $99 one-time →

340% avg prompt clarity gain

<200ms pipeline overhead

7 specialized stages

 Finder File Edit View

14:07

Works with

Claude

GPT-4o

Gemini

DeepSeek

Llama 3

Mistral

Ollama (local)

Under the hood

Before your words
reach the model.

Every request you submit runs this exact sequence — always, in under 200ms, invisibly. Click any stage to see the detail.

Intent Decomposer

NLP · Parsing

Parses your raw input into structured intent, verb, and desired outcome.

Your raw input is tokenized and analyzed to extract the primary verb, subject, object, and desired outcome. Intent Decomposer distinguishes between imperative instructions ("build X"), exploratory questions ("how does Y work"), and creative briefs ("write Z in the style of"). This classification routes your request to the appropriate assembly strategy in stage 5.

INPUT make a login endpoint

OUTPUT intent: CREATE | domain: backend-api | subject: authentication-endpoint | constraints: implied[security,error-handling]

Entity Extractor

NER · Pinning

Identifies and pins technical terms that must survive transformation intact.

Named entity recognition identifies and annotates all domain-critical tokens: proper nouns, version numbers, library names, acronyms, quoted strings, and technical identifiers. These entities are pinned with immutable flags so no downstream stage paraphrases, synonymizes, or drops them from the final output.

INPUT write a test for the UserAuthService.login() method

OUTPUT pinned: [UserAuthService] [login()] [test] — type: proper-noun, method-ref, test-directive

Complexity Classifier

Scoring · Routing

Scores the request across four dimensions and selects the optimal assembly strategy.

Four orthogonal dimensions are scored on a 0–1 scale: scope (narrow vs. sweeping), depth (surface vs. expert), ambiguity (clear vs. underspecified), and technicality (casual vs. domain-expert). The combined score selects one of 12 assembly strategies and determines how much context the Context Engine should inject in the next stage.

INPUT explain how transformers work

OUTPUT scope: 0.7 | depth: 0.8 | ambiguity: 0.4 | technicality: 0.6 → strategy: pedagogical-depth

Context Engine

Context · Memory

Injects system context: active app, clipboard history, recent prompts, and domain priors.

The Context Engine reads the macOS accessibility layer to determine your active application and selected text, queries the local clipboard history ring for recent content, and applies domain priors derived from your style profile and recent prompt chain. This context is injected as structured annotations into the assembly buffer in stage 5.

INPUT (active app: Xcode, selected: 42 lines of Swift, clipboard: UserDefaults key)

OUTPUT context injected: lang=swift, framework=swiftui, recently-edited=UserDefaults, scope=mobile

Prompt Assembler

Assembly · Templates

Constructs the final optimized prompt from classified intent, pinned entities, and context.

The Assembler takes all upstream data — classified intent, pinned entities, complexity score, injected context — and renders the final prompt using the active style template. The Engineering template adds role specification, constraint enumeration, format requirements, and test stub directives. The Research template adds academic framing. Each template is a composable specification, not a static string.

INPUT pinned entities + intent + context + style=Engineering

OUTPUT

You are a senior iOS engineer. Task: refactor the selected Swift code using SwiftUI best practices. Constraints: maintain UserDefaults key compatibility, add unit test stubs for each public method. Format: annotated code + change rationale.

Model Execution

Routing · Streaming

Routes the assembled prompt to the configured AI provider with streaming output.

The assembled prompt is routed to your configured endpoint — local Ollama, or a cloud provider via the optional PromptCraft Cloud proxy. Responses stream token-by-token into the output area with a blinking cursor. Timeout management, automatic retry on transient 5xx errors, and request deduplication are handled transparently. You see output in under 200ms from submission.

INPUT assembled prompt → configured provider (Claude / GPT-4o / Ollama)

OUTPUT streaming: ████████████░░░░░░░░ tokens received @ 45 tok/s, latency: 180ms

Post Processor

Validation · Cleanup

Strips model artifacts, validates format, and delivers clean output ready to copy.

The model response passes through format validation (is it actually code if code was requested?), artifact stripping (removes "Certainly! Here's..." preambles), verbosity enforcement (Concise mode truncates; Detailed mode requests continuation if output is too short), and structural cleanup. The final clean output is placed in the output area with copy, export, and clear actions immediately available.

INPUT raw model response with preamble and filler

OUTPUT cleaned: preamble stripped, format validated, verbosity constraint applied → output ready

Why it works

The AI doesn't need your
context. It already has it.

Live transformation

You type

Model receives

Seven optimization styles

Engineering adds role specification, constraints, format requirements, and test stubs. Research adds academic framing and source awareness. Content adds narrative structure and audience tone. Academic adds formal register and citation structure. Seven complete, composable templates — plus custom styles you define and save permanently.

Runs entirely local

The full 7-stage pipeline executes on your Mac. Your prompts never reach an external server until they touch your chosen AI provider — and with Ollama, that never happens either. Zero telemetry. Zero logging. The accessibility permission is read-only and sandboxed.

⌘⇧Space from anywhere

Lives in your menu bar. One global shortcut summons it from any app, on any Space, over any window. You type, optimize, copy — the whole cycle in under 10 seconds. No context switching to a browser tab. No login screen. No onboarding flow every session.

Context-aware by default

The Context Engine reads your active application, selected text, and clipboard history before assembling any prompt. Coding in Xcode? It knows the language and framework. Writing in Notion? It adjusts tone accordingly. The pipeline always knows what you were doing — without you having to say it.

Made by a developer, for developers

Oguzhan Atalay

@oguzhnatly · Lisbon, Portugal

"I got tired of starting every AI conversation with a three-paragraph context dump. I was spending more time writing the setup than writing the actual work. PromptCraft started as a one-file Swift script in my personal tools. When it started saving me forty minutes a day, I turned it into a proper app."

265

GitHub stars flutter_carplay package

12+

Products in flight across SaaS verticals

Years shipping full-stack engineer

Follow the build hello@promptcraft.app

2024 Q1

First Swift prototype: a menu bar app that prepended a role prompt to every request. Embarrassingly simple. Worked immediately.

2024 Q2

Added the style selector and context engine. Response quality improvement was obvious enough to show to other engineers.

2024 Q4

Built the full 7-stage pipeline, clipboard history, and cloud proxy for provider routing without API keys.

2025

Published. Still using it every single day. Still the only AI tool I actually keep open.

Pricing

$99 once. Or $12 a month.
No annual trap.

Pro is a one-time purchase you own forever. Cloud adds managed AI routing for people who want zero API key management. Both include the full pipeline.

Pro

Buy once. Own forever.

Stop writing context.
Start getting answers.

macOS 14 Sonoma or later. Apple Silicon and Intel. 14-day free trial — no account, no credit card. License key delivered instantly by email after purchase.

Download for Mac — Free View all pricing options

Before your wordsreach the model,the work is done.

Before your wordsreach the model.