Token usage and AI cost controls

When using Bring Your Own Key (BYOK) cloud AI providers (such as Anthropic Claude or OpenAI GPT-4o), managing token usage ensures predictable software expenses. Oyma provides built-in token monitoring and context window controls.

Understanding meeting token metrics

Meeting transcripts are naturally text-heavy:

• Speech Rate: Most professionals speak 130–160 words per minute.
• Words to Tokens: In English, 1,000 words correspond to approximately 1,330 tokens.
• 45-Minute Meeting: Produces ~6,500 words = ~8,700 prompt tokens.

Approximate cost per meeting (wholesale provider pricing)

• Local Ollama (Llama 3.2): $0.00 (100% Free on-device).
• Anthropic Claude 3.5 Sonnet: ~$0.025 to $0.035 per meeting summary.
• OpenAI GPT-4o: ~$0.020 to $0.030 per meeting summary.
• OpenAI GPT-4o-mini: ~$0.002 per meeting summary.

Built-in cost control features

In Settings › AI › Usage & Budgets, you can configure guardrails:

1. Context Window Capping: Restrict input prompt sizes to prevent runaway costs on multi-hour recordings.
2. Monthly Usage Counter: Displays total input/output tokens consumed in the current calendar month.
3. Warning Threshold: Displays a confirmation prompt before sending requests exceeding 25,000 tokens.

Minimizing token consumption

• Summarize Highlights Only: Generate summaries from selected transcript segments rather than full recordings.
• Use Local Whisper First: On-device Whisper creates the initial transcript for free, sending only text to the summary LLM.

Related guides

• BYOK setup in Bring your own AI API keys.
• Free local setup in Ollama local setup and model configuration.
• Prompt tuning in Prompt engineering for meeting summaries.