Free, open-source tools for Claude Code by Casper Schive. Install in one command, activate with your email, use forever. No subscriptions, no API costs.
Deploy Claude Code to a Linux VPS and control it from Telegram on your phone. Your Claude Code runs 24/7 with full Opus 4.6 and 1M context, even when your computer is off. Fix code, control your smart home, check deployments — all from your pocket. The skill includes a 10-second watchdog that auto-recovers from crashes, daily auto-updates, and daily session resets for fresh context. Requires a $5/mo VPS and a Claude Max subscription.
Does this cost extra beyond Max? No. It runs on your existing Claude subscription. Zero API costs.
What if it crashes? It auto-restarts. The watchdog checks every 10 seconds and recovers automatically.
Do I need to know Linux? No. The setup skill walks you through every step interactively.
Stop clicking through Fiken. Just tell Claude what to do — book a receipt, draft an invoice, check a balance. 67 tools covering the entire Fiken accounting API. Works with Claude Code, Cursor, Windsurf, and Claude Desktop. Human-in-the-loop confirms before writing anything, so it cannot mess up your accounting. Requires a Fiken account and any MCP-compatible editor.
Can it mess up my accounting? No. It asks for confirmation before any write operation. You approve everything.
Do I need to code? No. Copy one config line and you are running.
What if Fiken changes? We update the MCP. You get all updates free, forever.
Stop paying per-token API costs for OpenClaw. This proxy lets you use your Claude Max subscription to power all your OpenClaw agents — Opus 4.6 with 1M context for everything. A local proxy translates OpenAI-format requests into Claude CLI calls, registered as a custom provider that bypasses model validation.
Does Anthropic allow this? The proxy uses the Claude Code CLI with your authenticated Max subscription. It's your subscription, your CLI, your server.
Will it work with all OpenClaw agents? Yes. It registers as a custom provider called "claude-max" in openclaw.json. All agents can use it.
What models can I use? Claude Opus 4.6 (1M context), Sonnet 4.6 (1M context), and Haiku 4.5. All the latest models.
Based on Andrej Karpathy's autoresearch — the project with 73,000+ GitHub stars that proved AI agents can run research autonomously. Karpathy, former Director of AI at Tesla and co-founder of OpenAI, showed that you can write a strategy document, point an AI agent at it, and walk away while it runs hundreds of experiments overnight. We've adapted this into a one-command skill for everyday optimization: blog quality, SEO scores, page performance, and more. You define what to improve and how to measure it. The agent loops — modifies, measures, keeps improvements, discards regressions — without you in the chair.
What can I optimize with this? Anything with a measurable number: blog quality scores, Lighthouse performance, response times, bundle sizes. If you can measure it, the agent can optimize it.
Will it break my code? No. Every experiment is a git commit. If the result is worse, the agent reverts. Only improvements stay.
How is this different from Karpathy's autoresearch? Autoresearch is for ML model training on GPUs. Autoloops takes the same pattern and makes it work for any optimization task — no GPU required.
Built by Casper Schive. All products are free and open source. GitHub | YouTube | Telegram Community