# Cutting my AI subscription bill by 60% was surprisingly easy > Source: > Published: 2026-08-14 17:45:19+00:00 # Cutting my AI subscription bill by 60% was surprisingly easy ## The Shift to Local and Open Source The biggest drain on my wallet was paying for specialized writing assistants and basic research tools. I've moved almost all of that to a local setup. If you have a decent GPU, running a local LLM agent is a massive win for privacy and cost. I stopped paying for a dedicated AI writing tool and switched to a combination of a local Ollama instance and some custom prompt engineering. For those who aren't technical enough to host their own, using a unified API aggregator is the way to go. Instead of five different $20 subscriptions, I just pay for the tokens I actually use. ## Replacing the "AI Wrapper" Subscriptions I noticed I was paying for three different tools that all essentially just sent my data to GPT-4o or [Claude](/en/tags/claude/) 3.5 Sonnet. Here is how I restructured my stack: **AI PDF Analyzers:** I ditched the monthly subscription for a dedicated PDF AI. Now, I just dump the files into a large context window model.**AI Meeting Notes:** I stopped paying for the automated transcription service and started using a basic open-source Whisper implementation for my audio files.**AI Image Gen:** I moved from a monthly subscription to a pay-as-you-go credit system or local[Stable Diffusion](/en/tags/stable%20diffusion/).**AI Copywriting:** I replaced the "premium" templates with a few well-crafted system prompts in a free-tier LLM. ## A Practical Tutorial for the Switch If you want to start cutting costs, here is a simple deployment path to move from paid wrappers to a more sustainable AI workflow: 1. Install Ollama to run models like Llama 3 or Mistral locally on your machine. 2. Set up a frontend like Open WebUI so you have a [ChatGPT](/en/tags/chatgpt/)-like interface without the monthly fee. 3. For the tasks that require "frontier" power, use a pay-per-token API key rather than a flat monthly subscription. ``` # Example: Installing Ollama on macOS/Linux curl -fsSL https://ollama.com/install.sh | sh # Running a lightweight model for basic writing tasks ollama run llama3 ``` ## Is it worth the effort? It depends on how much you value "one-click" convenience. The paid tools handle the infrastructure for you, but you pay a "convenience tax." Moving to a custom AI workflow requires a bit of a deep dive into how these models actually work, but the cost savings are immediate. The real-world result for me wasn't just saving money—it was actually getting better results. When you stop relying on a tool's pre-set "templates" and start practicing actual prompt engineering, you realize the underlying models are far more capable than the wrappers let them be. Most "AI features" in expensive software are just hidden prompts that you can easily replicate for free. [Next Reproducing 2 →](/en/threads/6219/) [a library of Claude prompt techniques](https://tanyan888.com/), with plenty of directly applicable cases. ## All Replies （0） No replies yet — be the first!