Recommendations

The Recommendations page shows actionable suggestions generated from your usage data.

7 recommendation categories

1. Model Rightsizing

Suggests cheaper models for tasks where expensive models are overkill.

Example: “42% of your Opus sessions produce <500 tokens. Switching these to Sonnet would save ~$18/month.”

2. Prompt Patterns

Identifies recurring prompt weaknesses with specific fixes.

Example: “Your prompts in apps/dashboard/ rarely include file paths. Adding paths would improve PQ by ~2.0 and reduce correction turns by 30%.“

3. Context Management

Suggests when to use /compact or /clear based on your session patterns.

Example: “You average 35 turns before compacting. Compacting at 20 turns would save ~15% on token costs.”

4. Workflow Optimization

Recommends workflow changes based on efficiency analysis.

Example: “Using plan mode for tasks with 5+ file changes would reduce iteration by ~40%.“

5. Tool Usage

Suggests better tool selection based on your patterns.

Example: “You use Bash grep 12x/week when the Grep tool would be faster and more reviewable.”

6. Budget Optimization

Recommends budget settings based on your spending patterns.

Example: “Your 95th percentile daily spend is $4.20. Setting a soft cap at $5 would alert you on unusual days.”

7. Team Patterns

For team leads — identifies team-wide patterns and training opportunities.

Example: “3 developers average PQ <4.0. A 30-minute prompt writing workshop could save the team ~$200/month.”

Managing recommendations

Each recommendation has a status:

Status	Meaning
New	Just generated, not yet reviewed
Acknowledged	You’ve seen it and plan to act
Resolved	You’ve implemented the suggestion
Dismissed	Not applicable or not interested

Update status from the dashboard or via the API:

POST /v1/intelligence/recommendations/:id/status
{ "status": "resolved" }

Estimated savings

Each recommendation includes an estimated monthly savings in tokens and dollars, calculated from your actual usage data.