Project Usage & AI Activity

The Usage page gives you a project-level view of how AI models, compute environments, and sandbox executions are being consumed inside a single repository.

It answers practical questions such as:

• How much AI activity has this project generated?
• Which models are being used most?
• When did usage spike?
• Which workflows are driving cost?

Everything shown here is scoped to the current project (for example, Sid Lais Team / content-repo) and reflects activity from chats, AI Task Discovery runs, task execution, and sandbox-backed operations.

Overview Cards

At the top of the page, you’ll see four summary cards:

• Compute Cost
• Total Tokens
• LLM Cost
• Grand Total

These update based on the selected time range and give you a quick sense of scale before diving deeper.

Compute Cost

This represents the cost of execution environments used by agents. It includes sandbox-backed operations such as:

• Running AI Task Discovery
• Executing multi-step task workflows
• Applying file changes inside an isolated environment

For example, if you trigger AI Task Discovery on a large backend repository, compute cost may increase even if token usage remains moderate. That’s because the system is analyzing file structure, scanning code, and preparing structured outputs in a sandboxed context.

This metric helps you distinguish between:

• Model-heavy usage (reasoning and generation)
• Execution-heavy usage (analysis and sandbox operations)

Total Tokens

This shows the total number of tokens consumed across all model interactions in the selected period. It includes:

• Chat messages
• Task-based reasoning
• AI Task Discovery runs
• Code generation and refactors

For instance, if you run multiple structured discovery scans in one day, you may see a sharp rise in token count even if only a few tasks were completed. That indicates deeper analysis per invocation.

Teams often look at this metric after:

• Large refactors
• Repository onboarding sessions
• Heavy documentation generation cycles

LLM Cost

This isolates the cost attributed specifically to language model usage.

It reflects the cost of:

• Model reasoning
• Text generation
• Code suggestions
• Task planning steps

For example, if you primarily use chat-based workflows for documentation and small edits, LLM cost may dominate while compute cost remains low.

This helps you understand whether your usage pattern is:

• Interaction-heavy (chat and reasoning driven)
• Execution-heavy (sandbox and task workflows)

Grand Total

The Grand Total combines:

• LLM usage
• Compute cost
• Sandbox cost

It gives a single number representing total project usage for the selected time range. This is especially useful when reviewing overall usage during:

• Sprint retrospectives
• Client reporting cycles
• Internal AI adoption reviews

AI Model Usage (Trend Graph)

The AI Model Usage section shows token consumption per model over time (for example, over the last 7 days).

Each model is visualized separately, such as:

• xai/grok-4.1-fast-non-reasoning
• phala-gpt-oss-120b
• claude-sonnet-4.5

This helps you see:

• Which model is being used most
• When model usage increases or drops
• Whether experimentation with a new model changed usage patterns

For example:

If you switch from a smaller model to a higher-capacity reasoning model during a refactor week, you may see a clear upward trend in token usage for that specific model. This makes model-level decisions observable rather than abstract.

The time range filter (such as “Last 7 days”) allows you to correlate model usage with development events like feature releases or audits.

AI Line Edits (Contribution Heatmap)

The AI Line Edits section visualizes AI-generated edits across the calendar year in a heatmap format.

Each square represents a day. The intensity reflects the volume of AI-driven edits.

Below the heatmap, you’ll also see:

• Most Active Month
• Most Active Day
• Longest Streak
• Current Streak

This gives you a sense of development rhythm.

For example:

• During an intense migration month, you may see dense activity blocks across consecutive days.
• During maintenance phases, activity may appear sporadic.

This view is useful for:

• Understanding sustained AI usage patterns
• Reviewing how consistently agents are used
• Identifying bursts of heavy modification activity

You can toggle between different modes (All, LLM, Tasks) to understand whether activity came from conversational usage or structured task execution.

Usage Leaderboard

The Usage Leaderboard summarizes AI activity at the user level within the project.

It includes metrics such as:

• Accepted diffs
• Tasks completed
• Tokens used
• Cost
• Agent lines of code

For example, in a collaborative team setting, you can see which contributors are:

• Running more task-based workflows
• Generating higher volumes of model interactions
• Contributing more AI-driven code edits

This is not a performance metric. It’s an activity transparency view that helps teams understand how AI workflows are distributed.

Detailed Usage Log

The table below the summary sections provides a granular breakdown of individual usage events.

Each row includes:

• Author
• Mode (e.g., TRIGGER, LLM)
• Associated chat or task
• Model used
• Tokens (input / output / total)
• LLM cost
• Compute cost
• Sandbox cost
• Total cost
• Timestamp

This is especially useful when:

• Investigating a spike in token usage
• Auditing a specific AI Task Discovery run
• Understanding the cost impact of a particular model

For example:

If total usage jumps on a specific day, you can scroll the log and see whether:

• Multiple AI Task Discovery triggers were executed
• A large reasoning-heavy task was run
• A model switch increased token cost per interaction

Pagination ensures you can navigate historical usage without overwhelming the interface.

Exporting Usage Data

At the bottom of the Usage dashboard, you’ll find an Export CSV button.

Clicking this will export the model usage data into CSV format. This is commonly used when:

• Preparing usage reports for stakeholders
• Performing offline analysis
• Importing usage data into BI or analytics tools
• Sharing usage summaries with finance or operations teams

The export includes structured model-level usage details so you can analyze trends outside the platform if needed.

How Teams Typically Use the Usage Page

In practice, teams use the Usage page to:

• Monitor AI adoption across the project lifecycle
• Review usage after major development phases
• Understand model selection impact
• Investigate unexpected spikes in token or compute consumption
• Export structured usage data for reporting

Rather than interrupting development, the Usage page operates as a visibility layer. It makes AI-driven workflows measurable and traceable, while keeping planning and execution inside the core project interface.