Ubby uses a credit-based system to measure and bill for the AI computation your agents consume. Unlike traditional software where you pay a flat fee regardless of usage, Ubby's credit system ensures you only pay for what you actually use. Understanding how credits work helps you optimize costs, predict expenses, and make informed decisions about which AI models to use for different tasks.
This article explains the fundamentals of Ubby's credit system, what activities consume credits, how consumption is calculated, and how to monitor your usage effectively.
What are Ubby credits?
Ubby credits are the unit of measurement for AI computation on the platform. Every time one of your agents processes information using an AI model, it consumes credits based on the amount of text processed and the specific model used.
Think of credits like fuel for your agents. More complex tasks that process larger amounts of information consume more credits, similar to how driving a longer distance consumes more fuel. More powerful AI models also consume credits at a higher rate, just as high-performance vehicles consume more fuel per mile.
Credits provide a unified measurement across different AI models and providers. Rather than tracking usage separately for each model, you have one simple metric—credits—that applies consistently across your entire Ubby usage.
How credits relate to tokens
Under the hood, AI models work with tokens, which are fragments of text. A token might be a whole word, part of a word, or even just a punctuation mark. For example, the sentence "Understanding Ubby credits" breaks down into approximately 4-5 tokens depending on the specific model.
Ubby credits are calculated based on token consumption. Different AI models have different credit costs per million tokens, reflecting their varying computational requirements and capabilities. When you see a model's pricing listed as "4,500 credits per 1M tokens," this means that processing one million input tokens with that model costs 4,500 credits.
You do not need to think about tokens constantly, but understanding the relationship helps you grasp why some operations cost more credits than others. Processing a large document naturally requires more tokens than processing a short message, and therefore consumes more credits.
What consumes credits?
Every interaction with an AI model consumes credits. Understanding the main consumption categories helps you predict and control costs.
Agent execution
Whenever an agent runs and uses an AI model to process information or make decisions, it consumes credits. This is the primary driver of credit consumption for most users.
The amount of credits consumed depends on several factors:
Input tokens: The information you provide to the agent, including your instructions, any documents or data the agent needs to process, and the context the agent maintains from previous interactions in a conversation. Longer prompts and larger documents mean more input tokens and higher credit consumption.
Output tokens: The response the agent generates. A brief answer consumes fewer credits than a detailed, lengthy response. Output tokens typically cost more credits per token than input tokens because generating text requires more computation than processing it.
Model choice: Different AI models have vastly different credit costs per token. A powerful model like Claude Sonnet might cost 4,500 credits per 1M input tokens and 22,500 per 1M output tokens, while a lighter model like GPT-5-nano might cost only 75 credits per 1M input tokens and 600 per 1M output tokens.
Tool usage and integrations
When agents interact with external tools and services through MCP integrations, these interactions can also consume credits, though typically much less than the AI model processing itself.
The actual API calls to external services do not consume Ubby credits—those are billed separately by the service providers if applicable. However, the AI processing required to understand tool outputs and decide what to do with them does consume credits.
For example, if an agent retrieves a document from Google Drive, reads it, and extracts key information, the retrieval itself does not consume significant credits, but processing the document content through the AI model to extract information does.
Context and memory
AI models maintain context within a conversation, remembering what was discussed earlier. This context gets included as input tokens in subsequent messages, consuming credits.
In a long conversation, the context can grow substantial. If you have had a back-and-forth discussion with an agent across ten messages, each new message includes all that previous context as input tokens. This is why very long conversations gradually consume more credits per message.
Some models have larger context windows than others, meaning they can maintain more conversation history. While this capability is valuable, be aware that maintaining extensive context does have a credit cost.
How credit consumption is calculated
Understanding exactly how Ubby calculates credit consumption helps you predict costs and optimize usage.
The basic formula
Credit consumption for any agent execution follows this formula:
Total Credits = (Input Tokens × Input Credit Rate) + (Output Tokens × Output Credit Rate)
For example, if you use Claude Sonnet-4 (which costs 4,500 credits per 1M input tokens and 22,500 per 1M output tokens) to process a request with 50,000 input tokens and generate 5,000 output tokens:
Input cost: (50,000 ÷ 1,000,000) × 4,500 = 225 credits
Output cost: (5,000 ÷ 1,000,000) × 22,500 = 112.5 credits
Total: 337.5 credits
The calculation happens automatically, and you can see the exact credit consumption for each request in your usage logs.
Why output costs more than input
You might notice that output tokens consistently cost more credits than input tokens across all models. This reflects the computational reality of AI models.
Processing input text—reading and understanding it—requires computation but follows predictable patterns. Generating output text—creating new, contextually appropriate responses—requires significantly more computation because the model must evaluate many possible responses and select the best one at each step.
The ratio between input and output costs varies by model but typically ranges from 3x to 6x. More sophisticated models often have higher output multiples because they consider more possibilities when generating responses.
Model-specific pricing
Each AI model available in Ubby has its own credit pricing per million tokens. These prices reflect several factors:
Model capability: More advanced models with better reasoning, longer context windows, or specialized capabilities cost more credits. You are paying for superior performance.
Computational requirements: Some models require more processing power to run, which translates to higher credit costs.
Provider pricing: Ubby accesses models from various AI providers, each with their own pricing. Ubby's credit system normalizes these different pricing structures into a consistent credit-based model.
You can view the complete pricing for all available models in your Ubby dashboard under "Model Pricing." This transparency lets you make informed decisions about which models to use for different tasks.
Tracking your credit usage
Ubby provides comprehensive tools for monitoring your credit consumption, helping you stay within budget and identify optimization opportunities.
The billing dashboard
Your primary hub for credit tracking is the Billing & Usage page, accessible from your account settings. This dashboard shows:
Monthly credit usage: A visual representation of how many credits you have consumed during the current billing period versus your plan's allocation. You can see at a glance whether you are on track to stay within your limit or might need to adjust your usage or plan.
Current consumption: The exact number of credits used so far this month, updated in real-time as your agents execute.
Remaining credits: How many credits remain in your monthly allocation, along with when your credits will reset (typically at the start of each billing month).
Usage status: Whether your consumption is "Normal" or if you are approaching or exceeding your plan limits.
The dashboard provides a quick health check of your credit usage without requiring you to dig into detailed logs.
Daily usage logs
For more detailed analysis, Ubby maintains comprehensive daily usage logs showing exactly how credits were consumed. Access these logs through the "Usage Logs" section of your billing settings.
The logs organize your usage by day, showing:
Total credits consumed each day
Number of requests made
Which models were used
Breakdown of token consumption
You can expand any day to see individual requests, complete with timestamps, model used, input/output tokens, and credit cost for each request. This granular visibility helps you understand usage patterns and identify any unexpectedly expensive operations.
Understanding usage patterns
Reviewing your usage logs regularly reveals patterns that inform optimization:
Peak usage times: When do your agents run most frequently? Understanding this helps you plan capacity and predict costs.
Model distribution: Which models account for most of your credit consumption? Are you using expensive models for simple tasks that could use cheaper alternatives?
Per-agent costs: If you track which agents generated which requests (visible in the thread information), you can identify which agents are most expensive to run and whether their value justifies their cost.
Unusual spikes: Sudden increases in credit consumption might indicate an agent running more frequently than intended or processing more data than expected, warranting investigation.
Monthly credit allocation and resets
Understanding how credit allocation works helps you plan usage and avoid interruptions.
How monthly credits work
Each Ubby plan includes a specific number of credits per month. On the Free plan, you receive 10,000 credits monthly. On Pro plans, you receive credits based on the tier you selected—anywhere from 50,000 to 400,000+ credits monthly.
These credits are available for the entire billing month. You can use them at any pace—all at once, evenly throughout the month, or clustered around specific activities. The only constraint is the total monthly amount.
Credit reset timing
Credits reset at the beginning of each billing month, based on when you subscribed or last changed plans. If you subscribed on the 15th of a month, your credits reset on the 15th of each subsequent month.
When credits reset, any unused credits from the previous month do not roll over. Your allocation returns to the full monthly amount regardless of how much you used previously. This "use it or lose it" approach means there is no benefit to under-using your allocation—you might as well leverage your full credit pool each month.
The reset timing is clearly displayed in your billing dashboard, showing exactly when your next reset occurs and how many days remain in the current period.
What happens when you run out
If you consume all your monthly credits before the reset date, what happens depends on your plan and settings.
On most plans, you cannot execute additional agent requests once credits are exhausted. Your agents will not run until credits reset or you upgrade your plan. This prevents unexpected overages but means you need to monitor usage to avoid interruptions to critical automation.
Some plan tiers offer overage protection or the ability to purchase additional credit bundles mid-month. Check your specific plan details to understand your options.
The best practice is monitoring your usage pace throughout the month. If you notice you are on track to exhaust credits before the reset, you can either optimize your usage (switch to cheaper models, reduce unnecessary agent executions) or upgrade to a plan with more credits.
Estimating credit consumption
Being able to estimate how many credits an agent will consume helps you plan usage and choose appropriate models.
Simple estimation method
For a rough estimate of an agent's credit consumption:
Estimate input size: How much text will the agent process? A typical document page is roughly 500 words, which translates to about 650-750 tokens.
Estimate output size: How long will the response be? A paragraph is roughly 100-150 tokens, a page is 650-750 tokens.
Look up model pricing: Check the credit cost per million tokens for your chosen model.
Calculate: Apply the formula mentioned earlier.
For example, an agent that processes a 10-page document (roughly 7,500 tokens) and generates a 2-page summary (roughly 1,500 tokens) using Claude Sonnet-4:
Input: (7,500 ÷ 1,000,000) × 4,500 = 33.75 credits
Output: (1,500 ÷ 1,000,000) × 22,500 = 33.75 credits
Total: ~68 credits per execution
If this agent runs 50 times per month, it would consume about 3,400 credits monthly.
Factors that increase consumption
Several factors can cause actual consumption to exceed estimates:
Conversation context: In multi-turn conversations, previous messages are included as input each time, increasing token counts beyond just the new message.
System instructions: Complex agent configurations with lengthy system instructions add to input tokens on every execution.
Tool outputs: When agents use tools that return substantial data (like retrieving large documents), processing that data consumes additional tokens.
Model context management: Some models are more efficient at managing context than others, affecting token consumption in longer interactions.
Start with simple estimates, then refine based on actual usage logs showing real-world consumption for your specific agents and use cases.
Best practices for credit management
Effective credit management ensures you get maximum value from your allocation without unexpected interruptions or costs.
Monitor regularly
Check your credit consumption at least weekly, more frequently if running many agents or approaching your limit. The billing dashboard provides a quick status check, while usage logs offer detail when needed.
Regular monitoring lets you identify issues early—an agent running more frequently than intended, unexpected usage spikes, or pace that suggests you will exhaust credits before month-end.
Set mental budgets per agent
For important agents that run frequently, establish a rough sense of how many credits they should consume monthly based on their typical usage. When actual consumption deviates significantly, investigate why.
This does not require precise tracking, just awareness of normal ranges. An agent that usually consumes 500-600 credits daily suddenly consuming 2,000 credits warrants attention.
Understand your consumption patterns
After a few months with Ubby, you will develop intuition about your consumption patterns. You will know roughly what percentage of your allocation gets used by which activities, making it easier to plan and predict.
This understanding also helps when evaluating plan changes. If you consistently use 80-90% of your monthly allocation, you are efficiently utilizing your plan. If you regularly use only 30-40%, you might consider a lower tier to reduce costs.
Plan for growth
As you build more agents and expand your automation, credit consumption naturally increases. Factor this growth into your planning. If you are at 70% of your allocation now and planning to deploy three more significant agents next month, you will likely need to upgrade your plan proactively.
What next?
You now understand how Ubby's credit system works, what consumes credits, how consumption is calculated, and how to track your usage. This foundation prepares you for the next article, where we will explore the different AI models available in Ubby and how to choose the right model for each task to optimize both performance and cost.