117+ public repos. 12 stars between them — I counted. Most people would hide that number. I'm leading with it, because why I keep shipping anyway is the one thing worth knowing before you scroll — and I'll tell you at the bottom of the tools list.
Everything here is zero-dependency Python (stdlib only, MIT-licensed): no bloated dependency trees, no supply-chain surprises, no pip install roulette. Built for AI agents, LLM apps, and the developers who babysit them.
🔥 Four layers shrink your token bill. Here's which one saves the most.
tokenseive — The four: compress input, map your codebase (95–97% fewer tokens), cut model output — and shrink tool output by up to 93%, the layer that saves most.
pip install tokenseive. → try it
A different tool steps into this slot every day. Yesterday it was someone else.
Don't browse 117+ repos. Find your problem, take the tool:
| Your problem right now | The fix | What it does |
|---|---|---|
| 🧨 "My LLM returns broken JSON" | jsonsalvage |
Extracts & repairs JSON from messy LLM output — fences, trailing commas, truncation |
| 🕳️ "What even IS this codebase?" | repolens |
X-rays any repo in one command → interactive dependency graph + health grade in a single HTML file |
| 🔥 "I'm burning tokens" | tokenseive |
Multi-layer token optimization — compress prompts, map codebases, shrink output |
| 🧠 "My agent forgets everything" | mnemokit |
Persistent agent memory — SQLite + FTS5 full-text search, zero deps |
| 🗺️ "My agent reads whole files to answer one question" | codegraph |
Cached, incremental code-graph maps — agents query structure instead of reading code |
| 🔐 "Secrets keep leaking into prompts" | cloakpipe |
Redacts secrets & PII before the API call, restores them in the response. 16 detectors |
| 🎲 "My prompts are guesses" | prompt-bench |
Scores prompt quality on 12 dimensions, detects anti-patterns, A/B tests variants |
| 🔌 "I need an MCP server yesterday" | mcp-wrap |
Wraps any REST API into an MCP server in minutes |
| 🚧 "My agent needs guardrails" | ratify |
Contract layer for LLM agents — deterministic checks + LLM judgment |
| 🌍 "I want a dashboard, not a backend bill" | situroom |
Zero-backend global situation room — live earth intel, 100% client-side |
No sign-up wall, no "book a demo." These run in production, right now:
| Demo | What happens when you click it |
|---|---|
| 🔬 lens.bigwinner.work | Paste any public repo URL → get the repolens X-ray (dependency graph, blast radius, health grade, hotspot map) in your browser. No install |
| 🛍️ aiops.bigwinner.work | AI Store Optimizer — the Shopify app, embedded and running on a live storefront |
| 🏗️ bigwinner.work | The hub — every tool, guide, and live demo I ship, in one place |
🔍 repolensX-ray any codebase in one command — interactive dependency graph, blast radius, health grade, security scan & hotspot map in a single HTML file.
|
🗺️ codegraphCached, incremental code-graph maps so AI agents query structure instead of reading whole files. CLI + MCP — works with Claude Code, Codex, Cursor, Aider.
|
|
Multi-layer token optimization for LLM apps — compress prompts, map codebases, reduce output. Zero required dependencies.
|
Assemble a guild of AI agents that self-builds a roster, self-runs a task DAG, and self-grows a memory of lessons — offline, zero API keys.
|
📦 The full toolbox — 70+ more zero-dependency tools (click to expand)
🤖 AI agents & LLM infrastructure
| Repo | What it does |
|---|---|
talaria |
The agent runtime that ships — typed deterministic detection + phase-gated resumable pipelines |
guildhall |
🏰 A guild of AI agents that self-builds a roster, runs a task DAG, grows a memory of lessons — offline, zero API keys |
agent-loop-lite |
Simplified Plan→Act→Observe loop for building agents |
agent-memory |
In-memory + file-based multi-layer memory for agents |
signet-lite |
Lightweight persistent memory for CLI agents |
context-window |
Smart context-window manager with intelligent compression |
hermes-tool-guard |
Rate-limiting & safety middleware for LLM tool calls |
hookcatch |
Local webhook inspector — capture & inspect HTTP requests, nothing leaves 127.0.0.1 |
🛠️ Developer utilities
| Repo | What it does |
|---|---|
configkeeper |
Multi-format config loader |
quickenv |
Minimal .env loader with type casting & validation |
filewatch |
Cross-platform file/directory change monitor |
shellmate |
Safe shell-command execution wrapper |
texttools |
Text-processing utilities |
videoforge |
Automated short-video generation pipeline — script, voiceover, subtitles, composition |
Every one: Python stdlib only. MIT. pip install and forget.
I give the machines away and only sell the factory tours. Start here — it's genuinely enough for most people:
| Free thing | What you get |
|---|---|
⚡ dev-prompts |
120+ copy-paste AI prompts for engineers — code review, debugging, testing, refactoring. Dependency-free CLI |
| 🧰 ToolNest | 12 free browser tools — JSON formatter, Base64, password gen, QR & more. 100% client-side, your data never leaves the tab |
| 📚 Everything above | 117+ MIT-licensed repos. Fork them, gut them, ship them in prod. No attribution begging |
Why give this much away? Here's the loop from the top closed: every tool on this page runs daily inside my own autonomous AI agent.
Stars were never the business model — the tools pay me back in saved hours whether anyone notices or not. You're browsing my actual production stack, not a portfolio.
Honestly — don't buy anything here until the free shelf fails you. Most people's use case is covered by the free tools, and I'd rather you star a repo than refund a purchase.
Still reading? The store has the assembled versions — curated prompt packs, templates, and step-by-step guides that skip you past the assembly work the free repos leave you:
…and if a free tool already saved you an afternoon, coffee is the exchange rate:
A ⭐ costs you one click. It pays the next stranger who's stuck on the same problem and finds the tool because you starred it.
🏗️ bigwinner.work — the hub for everything I'm building · 🛍️ duc-store · ☕ Ko-fi