rafsilva85/credit-optimizer-v5

47% of your Manus credits are being wasted. This fixes it automatically.
> **Zero downsides. 47% average savings. Pays for itself in ~36 prompts.**
> Audited across 53 adversarial scenarios. Quality Veto Rule: if it can't save without loss, it doesn't touch your task.
**Get it:** Landing Page · Gumroad ($12) · Power Bundle ($12) · SkillFlow
**Free MCP Server:** PyPI · Smithery
As Seen On
Dev.to · PulseMCP · Smithery · MCPServers.org · Agensi.io · Indie Hackers · Awesome Manus AI
---
Why Your Credits Disappear
Manus charges credits per task. Most users waste **30–75%** because of four default behaviors:
| Problem | What Happens | Waste |
|---------|-------------|-------|
| **Wrong model routing** | Simple tasks run in Max mode when Standard produces identical results | Up to **5x overpay** |
| **Slow browser navigation** | Each page takes 8–45 seconds via browser tool calls, burning credits on wait time | **150+ seconds** per 10 URLs |
| **No chat detection** | Tasks that cost $0 in Chat Mode (Q&A, brainstorm, translation) run in Agent Mode | **100% overpay** |
| **Context bloat** | Tokens accumulate across steps — each subsequent step costs more than the last | **Exponential growth** |
These aren't bugs. They're default settings. The Power Stack overrides all four.
---
Zero Downsides
This is the most important thing to understand:
> **The Power Stack can never make things worse.** It has a hardcoded Quality Veto Rule: if an optimization would reduce output quality by even 1%, it skips that optimization entirely. In the worst case, it does nothing. In the best case, it saves you 75%.
There is no scenario where installing the Power Stack produces a worse result than not having it.
---
What's Inside
Credit Optimizer v5 — Smart Routing
Analyzes every prompt before execution and decides the cheapest path that delivers identical quality.
- **Intent Classification** across 12 categories (code, research, creative, data, translation...)
- **Model Routing**: Standard vs Max — only uses Max when complexity genuinely requires it
- **Chat Mode Detection**: Routes simple Q&A to Chat Mode ($0 cost)
- **Batch Detection**: Groups related tasks to reduce overhead
- **Context Hygiene**: Compresses accumulated context to keep token costs linear
Fast Navigation v2.0 — Raw Speed
Replaces slow browser tool calls with a programmatic toolkit.
- **httpx + selectolax** replaces Playwright overhead (30–2,000x faster)
- **Browser Cookie Bridge** for authenticated sites
- **Async parallel fetching** — 10 URLs in 1.3 seconds instead of 150+ seconds
- **Intelligent disk caching** with TTL — repeated visits are instant
Together: The Compound Effect
Credit Optimizer decides **what** to optimize. Fast Navigation decides **how fast** to execute it. Together, they multiply:
| Metric | Without Stack | With Stack | Improvement |
|--------|--------------|------------|-------------|
| Simple Q&A cost | Full price | $0 (Chat Mode) | **100% saved** |
| 10-URL research | 150+ seconds | 1.3 seconds | **115x faster** |
| Web scraping task | Full credits | 33% of original | **67% saved** |
| Full-stack web app | Full credits | 40% of original | **60% saved** |
| Research report | 12 minutes | 2 minutes | **6x faster** |
| Average across all tasks | Baseline | 47% less | **47% saved** |