Token limit message using Sidekick AI or the Query Builder Analyze

When you use the "Analyze using AI" button in the Query Builder, or Sidekick, with large amounts of data, you may see a "Request too large", "rate limit" or "tokens per minute (TPM)" error. Tokens are the units of text the model uses, and 1,000 tokens is roughly 750 words or pieces of data. This message comes from your AI provider (OpenAI or Anthropic) rather than from Ezekia, and your limits depend on your account tier with that provider. You can check your limits here:

• OpenAI: https://platform.openai.com/account/rate-limits
• Anthropic: https://platform.claude.com/settings/limits

This error occurs when the information being processed is too large for a single request. Try narrowing the data down first using the available filters, or the columns you include (selectable under the cog tick boxes), before asking your questions.

How to Avoid Data Limits

Here are some tips on using filters to manage data size and keep your questions focused:

1. Use Pre-Filters to Narrow Down People and Assignments
   • Focus on relevant criteria: If your query is about a specific role (e.g. “CEO”), use the Person title filter to display only entries matching that title. Or if you’re only interested in candidates in a certain location, narrow those people down first in your results.
   • Date filters: Use Note created at or Assignment last updated to restrict entries to recent records, which reduces the data size.
   • Assignments: For questions on assignments, use filters like Assignment status, Assignment client, or Assignment industries to view only the entries that match your needs.
2. Prioritise Key Fields
    Certain fields are more relevant to answering specific types of questions. For instance:
   • For leadership roles, focus on Person summary, Person aspirations, or Interview Notes.
   • If reviewing qualifications, use filters such as Person qualifications or Person positions.
   • For communication details, consider limiting to essential contacts in Person emails or Person Notes.
3. Use Targeted Tags and Statuses
   • Tags: For more detailed searches, use Person tags to filter for specific categories or attributes relevant to your query.
   • Current status: If you’re only interested in people actively looking for opportunities, apply the Person current status filter to refine results.
4. Simplify Questions and Compare Key Metrics
   • Ask focused questions: Instead of broad questions like "Who is the best CEO?", try more specific questions such as "Which CEO has the highest growth rate?".
   • Use Assignment details: For evaluating work history, the Assignment start date, Assignment end date, or Assignment compensation can provide focused information without extra details.
5. Chunk the data up
   • Use the begins with filter: If perhaps working with a large list of people, it might be worth adding the name filter and then the begins with operator and go alphabetically. Use an Or operator to go through several letters at once.

Still Seeing Issues?

By following these steps, you'll get the best results while staying within your AI provider's limits. For more information on those limits, contact your provider's support team.