[Feature]: Session-Level AI Usage & Cost Tracking

[vernonthedev]

Before submitting

• I searched existing issues and did not find a duplicate.
• I am describing a concrete problem or use case, not just a vague idea.

Area

Agent Runtime / Analytics / UI

Problem or Use Case

T3 Code supports multiple AI providers and models, but users currently have limited visibility into their actual usage during a session.

When working across different providers, it is difficult to answer questions such as:

• How many tokens have been consumed?
• Which provider is being used most?
• Which model generated the most responses?
• What is the estimated cost of the current session?
• How much usage came from background agents versus the active agent?

Without this visibility, users cannot effectively manage costs or understand resource consumption.

Proposed Solution

Introduce a session-level usage dashboard that aggregates usage metrics across all providers and agents.

The dashboard should track:

• Input tokens
• Output tokens
• Total tokens
• Request count
• Estimated cost
• Model usage breakdown
• Provider usage breakdown

Example UI

Session Usage

Provider          Requests     Tokens      Cost
------------------------------------------------
OpenAI            42           512,000     $3.12
Anthropic         18           214,000     $1.76
Google            9            67,000      $0.22

Total
------------------------------------------------
Requests: 69
Tokens: 793,000
Cost: $5.10

Additional Metrics

Per-agent statistics:

Agent A
- Requests: 24
- Tokens: 180k
- Cost: $1.12

Agent B
- Requests: 13
- Tokens: 96k
- Cost: $0.74

Per-model statistics:

Claude Sonnet
GPT-5
Gemini 2.5 Pro

Why This Matters

Benefits:

• Better cost awareness
• Easier provider comparison
• Improved budgeting
• Better visibility into agent behavior
• More informed model selection

MVP Scope

Track:

• Requests
• Input tokens
• Output tokens
• Total tokens

Display:

• Current session totals
• Provider breakdown
• Model breakdown

Estimated cost calculations can be added incrementally where provider pricing data is available.

Future Enhancements

• Historical usage analytics
• Daily/weekly/monthly reports
• Export usage data
• Team usage dashboards
• Cost alerts and budgets

Acceptance Criteria

• Usage is tracked per provider.
• Usage is tracked per model.
• Session totals are visible in the UI.
• Metrics update in near real time.
• Statistics reset appropriately when a session ends.