Sable Intelligence

→ vercel/next.jsscanned ✓·$NEXT generated·→ ggerganov/whisper.cppscanned ✓·$SPACES generated·→ supabase/supabasescanned ✓·$BASE generated·→ microsoft/TypeScriptscanned ✓·$TSCORE generated·→ langchain-ai/langchainscanned ✓·$CHAIN generated·→ facebook/reactscanned ✓·$HOOKS generated·→ openai/whisperscanned ✓·$VOCAL generated·→ shadcn-ui/uiscanned ✓·$SHADE generated·→ prisma/prismascanned ✓·$SCHEMA generated·→ vitejs/vitescanned ✓·$VIBE generated·→ vercel/next.jsscanned ✓·$NEXT generated·→ ggerganov/whisper.cppscanned ✓·$SPACES generated·→ supabase/supabasescanned ✓·$BASE generated·→ microsoft/TypeScriptscanned ✓·$TSCORE generated·→ langchain-ai/langchainscanned ✓·$CHAIN generated·→ facebook/reactscanned ✓·$HOOKS generated·→ openai/whisperscanned ✓·$VOCAL generated·→ shadcn-ui/uiscanned ✓·$SHADE generated·→ prisma/prismascanned ✓·$SCHEMA generated·→ vitejs/vitescanned ✓·$VIBE generated·

●The Problem

Every repo holds a product. Most builders never see it.

Developers ship tools that other developers build on top of. Every public repo is a signal — a stack, a capability, a narrative waiting to be extracted. The problem isn't access. It's depth.

Surface-level scanning returns surface-level results. What you need is a tool that actually reads the codebase — structure, patterns, dependencies — and reasons through what it finds before it writes a word.

skill patterns

30s

average scan

build ideas / repo

logins required

●Live Demo

Two layers of intelligence.

#1 Scan

Read any repo. Instantly.

Open any public GitHub repository. Sable reads manifest files, detects your stack, and maps 59 skill patterns against what it finds — in under thirty seconds.

→ no cloning · no uploads · server-side

→ reading manifest…

→ stack detected: TypeScript · Bun

→ 59 patterns matched ✓

# Generated by Sable

# CLAUDE.md — whisper.cpp

## Project Overview

C++ port of OpenAI Whisper.

Core: speech-to-text inference.

## Tech Stack

Language: C++17, Python

Build: CMake 3.17+

## Selected Skills (6/59)

cpp-inference, audio-processing,

cross-platform-build, wasm-deploy

#2 Mark File

A CLAUDE.md written by extended thinking.

Not pattern matching. Not template filling. The engine uses Claude Fable 5 to understand how your codebase actually works before writing a single line.

→ powered by Claude Fable 5

#3 Build Ideas

Three launch-ready concepts. Every scan.

Each idea comes with a domain, a pump.fun ticker, a value prop, and the build angle — derived directly from what the repo can do. Not generic. Specific to the codebase.

→ ship the site · launch the token · repeat

SpacesTranscribe$SPACES

Transcribe any Twitter Space in 60s using whisper.cpp as the engine.

~4 days

VoiceMark$VOICE

Voice memo → CLAUDE.md context. Every Claude Code user is a potential holder.

~3 days

PodcastBrief$BRIEF

Auto-clip alpha drops from crypto podcasts. Feed wallets, not just degens.

~5 days

●Intelligence Score

Deep repo analysis. In seconds.

MARK doesn't skim. It reads manifest files, scores your stack against 59 curated patterns, and builds a structured intelligence profile before writing a single word.

87/100Strong build signal

✓8/59 skills matched to detected stack

✓Architecture pattern clearly identifiable

✓3 viable pump.fun utility concepts generated

Score breakdown

Stack depth

8 / 59 skills

How many curated patterns matched

Architecture

clear pattern

Identifiable structure for CLAUDE.md

Build signal

3 ideas found

Viable pump.fun utility concepts

Reasoning depth

full pass

Extended thinking on codebase intent

●How it works

From tab open to build idea in thirty seconds.

01 / Scan

Read the manifest

MARK reads package.json, CMakeLists, Cargo.toml — whatever defines the stack. No cloning. No uploads. Server-side in under ten seconds.

02 / Score

Match 59 patterns

Your stack scores against a curated library of 59 skill patterns. The result is a ranked profile of what the codebase can do and how to work with it.

03 / Generate

Two outputs, one click

A production-ready CLAUDE.md drops into your terminal. Three pump.fun-viable utility ideas land in tab two. Rescan any repo, anytime.

●Features

Built for the crypto builder workflow.

Every feature is optimised for one thing: getting from "interesting repo" to "live utility on pump.fun" faster than anyone else.

Universal

Any public repo

GitHub URL is all you need. MARK works on repos in any language — C++, Python, Rust, TypeScript, Go — and doesn't require a token or install.

Instant

30-second full scan

Manifest fetch is parallel. Pattern matching is deterministic. Generation runs on the server. You get results in under half a minute.

Structured

CLAUDE.md ready to use

The output isn't a summary — it's a Claude Code context file. Drop it in any repo and Claude understands the stack immediately.

Actionable

3 launch ideas per scan

Each idea includes a domain suggestion, pump.fun ticker, value prop, and estimated build time. The angle is always derived from actual repo capabilities.

Free tier

5 scans daily, no card

Free to use. No credit card. No account. Rate-limited per IP at 5 scans per 24 hours so it stays accessible without burning API costs.

Portable

Download as ZIP archive

One click packages CLAUDE.md, matched skill files, and a setup guide into a ZIP. Everything Claude Code needs to hit the ground running.

●Powered by Claude Fable 5 Extended Thinking

Fable 5 thinking. Not a scan. A read.

Most repo tools skim the surface — file names, dependency counts, keyword matches. MARK uses Claude Fable 5— Anthropic's latest and most capable reasoning engine: a mode that thinks step by step, revisits its own conclusions, and builds deep architectural understanding before generating a single word.

The result is a CLAUDE.md powered by Fable 5 extended thinking that Claude Code actually uses — not a generic template, but a file that reflects how your specific codebase works and what your agent needs to contribute.

→ initializing Claude Fable 5 extended thinking pass...

→ Fable 5 analysis: whisper.cpp is a native C++ ggml port, not an API wrapper

→ adjusting build instructions: users must run 'make' first, not npm install

→ Fable 5 reasoning: generating 3 verified utility concepts based on Swift UI & C++ integration

Read any repo.
Find your next build.

Every repo holds a product. Most builders never see it.

Two layers of intelligence.

Read any repo. Instantly.

A CLAUDE.md written by extended thinking.

Three launch-ready concepts. Every scan.

Deep repo analysis. In seconds.

From tab open to build idea in thirty seconds.

Read the manifest

Match 59 patterns

Two outputs, one click

Built for the crypto builder workflow.

Any public repo

30-second full scan

CLAUDE.md ready to use

3 launch ideas per scan

5 scans daily, no card

Download as ZIP archive

Fable 5 thinking. Not a scan. A read.

Sable's 59 Skills.
Auto-selected.

Ready to scan?

Every repo holds a product. Most builders never see it.

Two layers of intelligence.

Read any repo. Instantly.

A CLAUDE.md written by extended thinking.

Three launch-ready concepts. Every scan.

Deep repo analysis. In seconds.

From tab open to build idea in thirty seconds.

Read the manifest

Match 59 patterns

Two outputs, one click

Built for the crypto builder workflow.

Any public repo

30-second full scan

CLAUDE.md ready to use

3 launch ideas per scan

5 scans daily, no card

Download as ZIP archive

Fable 5 thinking. Not a scan. A read.

Sable's 59 Skills. Auto-selected.

Ready to scan?

Sable's 59 Skills.
Auto-selected.