Choose the right context depth for your use case
This guide helps you pick the right context depth tier for your agent type, with signals for when to move to a higher setting.
Not every agent needs a high context depth. The right tier depends on how your content is structured and what kinds of questions your agent handles. This article maps each tier to the use cases that benefit most.
The core question
How much of your content does the agent need to assemble a good answer?
Some questions have a single, focused answer. Others require pulling pieces from across your content, or remembering a long exchange. The first works well at Compact. The second needs more room.
Tier recommendations
Compact - FAQ and deflection bots, high-volume simple support. Best when questions have focused, self-contained answers and query credit cost matters most.
Balanced - General product and support agents that need slightly more grounding. The everyday default for most agents.
Extended - Documentation and technical assistants over larger knowledge bases, or agents in longer multi-turn conversations that need to hold the thread.
Rich - Research assistants and complex B2B sales agents answering layered questions across a large content set while keeping the full conversation in view. If you also use a premium model, costs increase further - see How context depth affects query cost.
Max - Compliance, legal, and financial review where completeness is more important than cost. Pairs well with Verify Responses for the highest-assurance answers. Note that Max + Verify Responses is the highest-cost combination available, and using a premium model adds further cost - review How context depth affects query cost before committing to this combination.
Custom - Enterprise accounts with specific context and history requirements set by their account team.
Signals that you need a higher setting
Move up one tier when you see any of these:
| What you observe | What it means |
|---|---|
| The answer is in your content but the agent doesn't always find it | The agent isn't reading far enough into your content |
| The agent's answer is incomplete - it misses details you'd expect | The full answer is spread across multiple places and the agent isn't reading all of them |
| Users ask questions that require comparing or summarizing several things at once | The agent needs to hold all of them in view at the same time to answer well |
| Users ask broad questions that cover many topics in your content | The agent is only reading a portion of the relevant content for that question |
| The agent forgets what was said earlier in a long conversation | The agent's conversation memory isn't long enough to keep up with the thread |
| Your content has many similar sections and the agent keeps giving the same answer | Similar sections are crowding out other relevant content the agent should also consider |
Signals you do not need a higher setting
- The agent answers correctly and consistently at the current depth
- Questions are focused and answered from one section of your content
- Conversations are short with no need to reference earlier turns
Starting guidance
Start at Balanced (the Premium default) and raise only when a specific problem arises. On Standard, Compact is the only available option.