aura tokens

Count tokens in a file or stdin.

Syntax

aura tokens [flags] <file>
cat file | aura tokens [flags]

Description

Counts tokens using the configured estimation method from features/estimation.yaml. Reads input from a file path argument or stdin.

Multiple estimation methods can be specified for comparison. When a single method is used, output is a plain number. When multiple methods are specified, output is labeled (method: count).

Native estimation uses the current agent’s provider. Use --agent, --provider, or --model root flags to control which provider is used, or omit to use the default agent.

Delegates to the built-in Tokens tool — the same tool the LLM uses via aura run.

Flags

Flag	Default	Description
--method	from config	Estimation method (repeatable): rough, tiktoken, rough+tiktoken, native

Plus all global flags.

Methods

Method	Description
rough	Character count divided by divisor (default 4)
tiktoken	Token count via tiktoken encoding (default cl100k_base)
rough+tiktoken	Maximum of rough and tiktoken estimates
native	Provider’s native tokenization using the resolved agent’s provider

Examples

# Count tokens in a file (uses configured default method)
aura tokens path/to/file.go

# Count tokens from stdin
cat large-prompt.txt | aura tokens

# Override estimation method
aura tokens --method tiktoken path/to/file.go

# Compare multiple methods
aura tokens --method rough --method tiktoken path/to/file.go

# Native estimation (uses default agent's provider)
aura tokens --method native path/to/file.go

# Native with explicit provider/model
aura --provider ollama --model qwen3:32b tokens --method native path/to/file.go

# All methods at once
aura tokens --method rough --method tiktoken --method rough+tiktoken --method native path/to/file.go