Skip to content

LLM AI Providers

Jump to bottom
Nick Bradshaw edited this page Dec 26, 2025 · 3 revisions

LLM AI Providers & API Keys

Scribble relies on a Large Language Model (LLM) to analyze your transcripts and generate summaries. We currently support four major providers. This guide explains the differences, costs, and how to get set up.

🪙 Understanding Tokens & Cost

Most AI providers charge based on Tokens, not requests.

  • What is a Token? Think of a token as a piece of a word. Roughly 1,000 tokens 750 words.

  • Input vs. Output:

  • Input Tokens: The text you send to the AI (your transcript + your system prompt). This is usually cheaper.

  • Output Tokens: The text the AI writes back to you (the summary). This is usually more expensive.

  • The Cost Reality: If you upload a 1-hour transcript, that might be 10,000 to 15,000 tokens of Input. The summary might be 500 tokens of Output. You are billed for both. Your LLM AI provider of choice should have listed their cost per million tokens. Even at 20,000 tokens, you're probably paying pennies, nickles, or dimes per recap. But that's just a wild estimate, and your experience may be different.

1. Google Gemini

Google's Gemini is currently the most accessible provider for personal projects due to its generous free tier.

  • Free Tier: Google offers a "Free of Charge" tier for their Flash models (e.g., gemini-2.5-flash, gemini-3-flash-preview).

  • Limitation: You cannot access "Pro" or models on the free tier.

  • Privacy: Data sent to the Free tier may (meaning, probably will) be used to train Google's models.

  • Paid Tier: Pay-as-you-go. Access to higher powered (Pro) models. Your data is not used for training.

  • Live Tested: ✅ Yes. Confirmed working on v3.0.0.

How to get a Google API Key

  1. Visit Google AI Studio.
  2. Sign in with your Google Account.
  3. Click "Get API key" in the top left.
  4. Click "Create API key" (you may need to link it to a Google Cloud project).

2. Anthropic (Claude)

Anthropic's Claude is known for high-quality, human-sounding writing and excellent instruction following.

  • Cost: Strictly Pay-as-you-go. You must pre-load your account with credits (min $5).
  • Models: claude-sonnet-4-5 could be considered the state-of-the-art for reasoning and writing.
  • Live Tested: ✅ Yes. Confirmed working on v3.0.0.

How to get an Anthropic API Key

  1. Visit the Anthropic Console.
  2. Sign up and verify your email/phone.
  3. Go to Settings > Billing and add credits.
  4. Click "Get API Keys" and create a new key.

3. Ollama (Self-Hosted / Local)

Ollama allows you to run open-source models (like Llama 3, Deepseek-r1, Mistral, etc.) on your own hardware.

  • Cost: Free (hardware dependent).
  • Live Tested: ✅ Yes. Confirmed working on v3.0.0.

⚠️ Important Considerations for Ollama

  1. Hardware Requirements: Running an LLM is heavy. If you send a 20,000-token transcript to a local model, your server needs significant RAM/VRAM. If your hardware is weak, Scribble is likely to time out waiting on a response from Ollama (Which will continue to run and consume resources, despite the request having timed out).
  2. No Attachments: The standard Ollama API does not support file attachments. Scribble automatically handles this by converting your transcript into raw inline text before sending it. This works fine but consumes "Context Window" space faster.

How to configure Ollama

Typically, you do not need an API key for Ollama.

  1. Ensure Ollama is running (ollama serve).
  2. Set LLM_PROVIDER to ollama.
  3. Set OLLAMA_URL to your server's address (e.g., http://192.168.1.50:11434 -- Note that if Scribble and Ollama are running on the same Docker network, you can use container name resolution to resolve the address, rather than an IP address).
  4. Set LLM_API_KEY to SomeMadeUpValue (Scribble requires a value here to pass validation, but it is ignored).

4. OpenAI (ChatGPT)

OpenAI's ChatGPT has been an industry standard model provider.

  • Cost: Strictly Pay-as-you-go.
  • Live Tested: ❌ Untested.
  • Note: While the code implementation follows the official OpenAI spec, I haven't purchased an API key to verify it live. Proceed with caution.

How to get an OpenAI API Key

  1. Visit the OpenAI Platform.
  2. Sign up and go to Settings > Billing.
  3. Add a payment method and purchase credits.
  4. Go to API Keys and create a new secret key.

Clone this wiki locally