claude code grátis pra sempre

Main concept / product

• NineHouter is a local or self-hosted proxy/router that lets you connect multiple LLM providers, including Gemini, OpenAI/OpenAI-compatible setups, “Cloud Code”, OpenHouter, and various Chinese model/tool providers.
• It can generate an API key and endpoint URL and is marketed as plug-and-play, with API compatibility that resembles Anthropic and OpenAI interfaces.
• It can be installed on a local machine or a VPS, with a setup flow that includes a dashboard / click-to-deploy option (the demo shows deploying via a hosting provider).

Key feature: multi-provider token rotation / failover

A core feature is automatic rotation:

• When one provider’s token limit runs out, NineHouter passes requests to another provider/account.
• The video describes the behavior as similar to an old floppy disk drive—cycling through configured sources.
• It also mentions different fallback modes, such as:
  • Try in order: sequential retry across configured providers.
  • Silent fallback: if one model fails, it tries the next while maintaining conversation continuity/shared context.

Model mapping / routing

Inside NineHouter, you can configure:

• The base model (the example mentions Cloud Code initially tied to a model type like “Hik 4.5”).
• Model routing/mapping, so calls to one model are translated to another (e.g., routing to GLM 5.2 or other models).

This enables combining providers while presenting a consistent interface to the client application.

Token/compression “token saver” options

The dashboard includes multiple options aimed at reducing token usage:

• RTK / Output compression: reduces tokens by cleaning command/noise before sending to the model.
• Conversation compression: compresses long sessions to avoid context explosion and loss of earlier context.
• “Caveman skill”: a style/strategy option (light/full/ultra) intended to reduce tokens (the speaker claims it can save many tokens).
• Additional modes are mentioned, such as a “lazy senior dev” style (not fully defined in the subtitles).

Observability / cost & token accounting

NineHouter provides visibility into:

• Input tokens
• Output tokens
• Estimated spending
• Which provider/model was actually used (including distribution across providers)

Tutorial/testing walkthrough (practical guide)

The video demonstrates this workflow:

1. Deploy NineHouter (click-to-deploy).
2. Configure a password/dashboard.
3. Use the endpoint + key to obtain an API URL and API key.
4. Paste the key + base URL into the “Cloud settings” page.
5. Use an example Ruby app (the video references a “Ruby API”) to confirm the app works without directly logging into the provider.

Example prompt / program described

• Build a Ruby on Rails 7 “rental company” API with SQLite, pass tests, and run end-to-end.

Results reported (analysis of token savings)

The speaker reports mixed/negative results for their specific task:

• With RTK/output compression enabled:
  • The run took much longer (around 7 minutes cited)
  • Token counts still appeared very high in the shown results
• With other options (including “caveman”):
  • They reportedly spent even more tokens

Even so, they recommend trying—suggesting outcomes may vary by use case.

Caution / risks (review/analysis)

The video frames risks like this:

• It argues the setup is “not piracy/illegal”, while acknowledging it may be “immoral.”
• It warns major providers (e.g., Gemini / Cloud Code) may detect proxy/tool usage and block accounts.

Recommendation framing

• Good for: students, side projects, and experimentation
• Not recommended for: fully trusting for daily production with real workloads
• Best used to test “Chinese models” cheaply/“for free,” but with risk

Overall: treat it as experimentation infrastructure rather than something you can safely rely on long-term for production.

Hosting and convenience

• The speaker mentions using Hostinger for deployment/testing and provides a coupon link.
• The video highlights the convenience of centralizing access via hosting rather than installing and managing everywhere.

Main speakers / sources (as referenced)

• Primary speaker: a YouTube creator (referred to as “I” throughout the narration).
• Coupon/code references include HostG/Hostinger and a name like Davin.
• Providers mentioned as sources:
  • Gemini
  • OpenAI / OpenAI-compatible
  • Anthropic (“Claude”)
  • Several “Chinese” model ecosystems (e.g., GLM 5.2)
• Later recommendation references an external YouTuber: “PewDiePie”.