Summary of "Claude Code Is Now 100% Free - Here's How"

Summary (tech setup + how-to + build tips)

Create a local project config
- In VS Code, create a project folder (named .claude—the subtitles show .cluade, likely meaning .claude).
- Add a local settings file: settings.json.local.
- This config is used to set up Claude Code locally per project.
Get an OpenRouter API key
- Go to openrouter.ai, create an account, then create an API key (e.g., “Claude code free”).
- Paste the API key into Claude Code’s local configuration (the authentication token field) and save.
Select a free coding model via OpenRouter
- In OpenRouter, browse models and search for “free”.
- Pick a free coding-suitable model (the subtitles mention NVIDIA models, including MiniMax M2.5).
- Paste the chosen model details into Claude Code’s configuration.
Set the Base URL
- Set the Base URL to OpenRouter’s endpoint (the exact value is implied and said to be provided later in the video).

Run Claude Code from the terminal.
Confirm the selected model (e.g., MiniMax free / MiniMax M2.5) is being used.
Check OpenRouter activity/logs to confirm the cost is $0, indicating the free tier is working.

The speaker emphasizes that Claude Code works best when you design your workflow and prompts correctly—especially with smaller/free models.

Use small, focused files instead of one huge file.
Why: the AI has a context window—stuffing everything into a single file can cause loss of context, more mistakes, and hallucinations.
Suggestion: keep files under ~600 lines.
Also suggested: store reusable rules/prompts in a Claude Markdown file so you don’t have to repeat them every time.

The AI won’t protect you from problems you don’t know about.
Example: if you need multi-user updates, you must think about race conditions—if you don’t know that concept, you won’t ask for it and bugs may appear under concurrency.

Break work into a task list (the subtitles mention a Plan Markdown file).
For each AI session:
1. Do one feature/task
2. Manually test
3. Push to GitHub
Benefit: if the AI breaks something later, you can revert/rebuild from a known-good commit.
This is described as especially important with free/smaller models, which may be less reliable.

The speaker references a completed app “CardioFlow” (a cardio workout generator) built in about an hour.
A link is mentioned in the description to try it.
The “five prompts” used to build CardioFlow are also said to be included in the description.

Main speaker: “Hasan” (author behind learnwithhasan.com and a “solo builder course”)
Sources/services mentioned:
- Claude Code (VS Code integration)
- OpenRouter.ai (API keys, model selection, logs/cost)