Input Length Limit Guardrail

The Input Length Limit guardrail (input_length_limit) blocks inputs that exceed a configurable length limit. It supports both character-based and token-based counting, and can evaluate either the most recent user message or all user messages in the conversation.

When to Use

Use this guardrail to: - Prevent users from sending excessively long prompts that could result in high API costs. - Avoid exceeding model context windows or token limits. - Enforce strict input limits on chat forms.

Configurable Fields

Field	Key	Type	Default	Description
Maximum length	max_length	Number	5000	The maximum allowed length. Interpreted as characters or tokens depending on the counting method.
Use token-based counting	use_tokens	Checkbox	false	When enabled, the limit is applied to the number of tokens instead of characters. Uses the ai.tokenizer service.
Tokenizer model	tokenizer_model	Textfield	gpt-4	The model to use for token counting (e.g., gpt-4, gpt-3.5-turbo). Only visible when token-based counting is enabled.
Scan all user messages	scan_all_user_messages	Checkbox	false	When enabled, every user message in the conversation is combined and measured together instead of just the most recent one.
Violation message	violation_message	Textarea	See below	The message displayed when the limit is exceeded. Supports placeholders: @count (actual length), @max (configured limit), @unit (characters/tokens).

Default Violation Message

Your input has @count @unit, which exceeds the maximum of @max @unit.

Example Configuration

Below is an example configuration for enforcing a 1,000-token limit across all user messages in the conversation using the GPT-4 tokenizer:

id: input_length_limit
max_length: 1000
use_tokens: true
tokenizer_model: gpt-4
scan_all_user_messages: true
violation_message: "The chat history is too long (@count @unit). Please start a new session or keep it under @max @unit."