Video summary

The ONLY Setup Guide You'll Need — Claude Code, Codex & Antigravity FREE (Full Tutorial)

Main summary

Key takeaways

Technology

Tech/Setup Summary (Claude Code, Codex, AntiGravity via “Smart AI Gateway”)

What the tutorial claims to build

  • A “Smart AI Gateway” using Nine Router to connect to multiple free AI providers.
  • The gateway exposes a single “combo” routing endpoint that:
    • Tries top-priority models first
    • Automatically falls back to the next available provider/model when a model runs out of quota/credits/window
    • Requires no manual switching
  • That one endpoint is then used to configure:
    • Claude Code / Claude Code Terminal
    • Claude Code Desktop
    • Codex (Codex CLI)
    • AntiGravity

Step-by-step workflow and key configuration details

1) Prerequisites installation

The video instructs viewers to install (roughly in this order):

  • Node.js
  • Git
  • Python
  • VS Code
  • Claude Code Terminal
  • Claude Code Desktop
  • Codex CLI
  • AntiGravity

Notes:

  • Commands differ for Windows PowerShell vs macOS
  • Verification commands like --version / -h are mentioned

2) Install and open “Nine Router”

  • Install Nine Router from a GitHub repo
  • Run it via CLI and open it in a browser
  • It may request an initial default password

3) Add multiple AI providers (free-trial style) and validate/free-model support

Inside Nine Router → Providers, the tutorial connects multiple sources and tests whether models are “working” (keeping only working models).

Provider examples mentioned

  • Kero AI / AWS Builder ID-based accounts

    • Video mentions a multi-Gmail-account method to obtain more free trial credits
    • Includes cookie clearing steps to re-auth with different Google accounts
    • Mentions credit examples like “50 credits per free trial account”
    • Notes a way to get a higher credit tier (requires a debit card) and warns about canceling to avoid charges
    • Models added include ones described as “OpenAI compatible” style (e.g., “Big Pickle”, “DeepSeek V4 Flash”)
    • Models that return errors (e.g., “internal server error”) should be deleted

  • Gemini (Gemini CLI / Google account authorization)

    • Multiple models are tested; failing ones are removed
    • Emphasizes “Gemini 2.5 Pro” as a priority candidate
    • Mentions a quota tracker and adding more Google accounts if quotas run out

  • “Kimchi” provider

    • Requires login and 2-step verification using an authenticator app
    • Tests models and keeps only those that work (example shown: Minimax M3 while others fail)

  • NVIDIA “API key” provider

    • Uses an API key and validates available models
    • Keeps models described like DeepSeek V4 Pro / DeepSeek V4 Flash; deletes the rest
    • Mentions GLM 5.2 not working and removing it

  • OpenRouter

    • Uses an OpenRouter API key
    • Tests free models; removes ones that are rate-limited or failing
    • Highlights a coding attempt like “Queen 3 Coder Free”, but it may fail due to temporary errors
    • Chooses whichever free model successfully validates

  • Google AI Studio (Gemini API)

    • Create an API key in Google AI Studio and add it to Nine Router
    • Validates model availability (example mention: 2.5 Pro)

  • “Arq/ArrowLink” custom provider (Cloudflare-like / free window credits)

    • Uses an OpenAI-compatible endpoint configuration:
      • Base URL
      • API key
      • Default model
    • Mentions a window period of ~5 hours and free credit usage
    • Validates models like “Cloud Ops / Cloud Fable 5” (as described), keeping working ones

  • Free models provider via OpenAI-compatible format endpoint

    • Mentions a custom free model source where GPT 5.5 can be used via an OpenAI-compatible endpoint
    • Notes that some endpoints (e.g., Anthropic format) require topping up
    • Therefore focuses on the OpenAI-compatible route for GPT 5.5

  • Potential other provider

    • Optionally add “Nara Router” (claims “5 million tokens daily”)
    • Tutorial suggests skipping if enough powerful models are already available

4) Build the “combo” (routing + priority + automatic fallback)

In Nine Router → Combos:

  • The combo name must follow a specific convention (video repeatedly suggests keeping it like “Cloud Opus free”), or integrations may fail
  • The combo defines an ordered priority list of models

Routing behavior described:

  1. Try the top-tier model from the first provider (example mentions possible priority like Cloudflare/ArrowLink Cloud Sonnet 4.5 Thinking + agentic coding if supported)
  2. If quota/window ends → fallback to the next provider/model (example: Kero AI model(s))
  3. Then fallback to free GPT 5.5 (OpenAI-compatible free endpoint)
  4. Then fallback to Gemini 2.5 Pro and other tested models in sequence

The gateway automatically handles quota exhaustion and fallback logic.

5) Integrate with “Claude Code” / Claude Code Terminal

In Claude Code CLI/tools:

  • Select the created combo in Nine Router
  • Ensure the combo naming (e.g., “Cloud Opus free”) matches what the tutorial expects

The tutorial shows:

  • Copy/paste configuration into a settings.json file (or apply via UI)
  • Run the terminal and issue a prompt to verify responses

Troubleshooting note (important):

  • If a provider doesn’t support a specific model (example: “Cloud Fable 5”), fix it by:
    • Removing that model from the combo
    • Replacing with an alternative such as Cloud Ops / Cloud Opus 4.8
    • Reordering for priority
  • After correction, console/logs show which model was actually routed/used (example: routing to “Sonnet 4.5…” then later “Cloud Opus 4.8” after the fix)

6) Add “Auto Harness” plugin to Claude Code

The video recommends an Auto Harness plugin:

  • Described as a “self-learning skills layer”
  • Learns from repeated coding sessions and turns recurring work into reusable skills
  • Setup is said to take ~30 seconds (install command from its GitHub repo)

Also includes steps for Claude Code Desktop:

  • Enable Developer Options
  • Configure the third-party interface using the Gateway base URL + auth scheme
  • Restart and select the created combo

7) Integrate with Codex (Codex CLI)

In Codex CLI:

  • Select models, with the tutorial stressing that GPT 5.5 is the most compatible/works best

Configuration steps:

  • Apply settings via UI or manually edit Codex config in a Codex folder (e.g., a config file)

Verification:

  • Run Codex commands (e.g., codex and a “hi” prompt)
  • Confirm it uses GPT 5.5 through the gateway

8) Integrate with AntiGravity

Requirements:

  • Run Nine Router as administrator
  • Configure AntiGravity setup under MITM tools

DNS/registration step:

  • Tutorial instructs adding DNS records by pasting a command that edits a host/registration file (Windows flow shown)
  • Start the server and verify it’s running

Then:

  • Select the combo model so AntiGravity routes traffic through the same gateway
  • Confirm by observing logs that show the request reaching the combo and which provider/model responded

Review/guide/tut highlights (what the tutorial emphasizes)

  • Watch every step: missing parts breaks the setup
  • Validate models (“check” to confirm working models) and remove non-working ones
  • Use quota-aware fallback via combo priority to avoid manual switching
  • Combo naming conventions (e.g., “Cloud Opus free”) matter for integration success
  • Includes a troubleshooting loop:
    • If a provider/model doesn’t support your desired model → remove it → add an alternative (e.g., Cloud Opus 4.8) → reorder → verify logs

Main speakers/sources (as identifiable from the subtitles)

  • Primary speaker: the tutorial narrator (“guys… let me show…”, “I will tell you”, step-by-step instructions)
  • Sources referenced by name:
    • Nine Router (GitHub repo / “Getup Repo”)
    • Claude Code (Terminal & Desktop)
    • Codex CLI
    • AntiGravity
    • Provider platforms mentioned: AWS Builder ID / Kero AI, Gemini, Kimchi, NVIDIA, OpenRouter, Google AI Studio, ArrowLink/Cloud models, plus a “free models” OpenAI-compatible endpoint provider
    • Plugin mentioned: Auto Harness (GitHub repo)

Original video