Token Usage

Tokens are the units AI models use to process text. Understanding tokens helps you optimize costs.

What Are Tokens?

Tokens are pieces of words:

"Hello" = 1 token
"Hello, world!" = 3 tokens
"implementation" = 1 token
"implementations" = 2 tokens

Rule of thumb: 1 token ≈ 4 characters or ¾ of a word.

What Consumes Tokens

Input Tokens (sent to AI)

Ticket title and description
Code context (relevant files)
Conversation history
Agent system prompt
Organization context

Output Tokens (generated by AI)

Clarifying questions
Implementation plans
Generated code
PR descriptions
Comment responses

Typical Usage

Task	Input Tokens	Output Tokens
Simple clarification	2,000	500
Small feature	10,000	3,000
Medium feature	25,000	8,000
Complex feature	50,000+	15,000+
PR feedback response	5,000	1,000

Reducing Token Usage

Write Concise Tickets

❌ Verbose (more tokens):
"I would like you to please add a feature that allows
users to upload their profile pictures to the system..."

✅ Concise (fewer tokens):
"Add user avatar upload: JPG/PNG, max 5MB, store in S3"

Keep Context Focused

Only include relevant information in your agent context:

❌ Too much:
[Entire architecture document, all coding standards,
company history, team bios...]

✅ Focused:

- Stack: NestJS, Drizzle, PostgreSQL
- Pattern: Repository for DB, DTOs for API
- Testing: Jest, required for new endpoints

Use Appropriate Models

Task	Recommended Model
Quick questions	Haiku
Standard features	Sonnet
Complex refactoring	Opus

Haiku is ~10× cheaper than Sonnet for simple tasks.

Break Up Large Features

Instead of one massive ticket, create smaller focused tickets:

❌ One large ticket:
"Build complete user management system"

✅ Multiple focused tickets:

1. "Add user CRUD endpoints"
2. "Add user avatar upload"
3. "Add user role management"
4. "Add user invitation flow"

Monitoring Usage

Dashboard

View token usage in Settings → Usage:

Daily/weekly/monthly breakdowns
Per-agent consumption
Per-project consumption

Per-Run Details

Each agent run shows:

Input tokens used
Output tokens generated
Total cost
Model used

Cost Estimation

Before running a large task, estimate tokens:

Count characters in your ticket + context
Divide by 4 for approximate input tokens
Estimate output as 20-50% of input
Apply your tier multiplier