Getting started

PCU Distribution

Understand how Paragon Compute Units are distributed across the platform.

Paragon Compute Units (PCUs) are the unit of measurement for work performed by Paragon. Every action Paragon takes — reviewing a PR, running tests, generating code, or executing an agent session — consumes PCUs based on the complexity of the task.

What Consumes PCUs

ActionPCU CostDescription
PR ReviewsVaries by PR sizeAutomatic code review on pull requests
Test GenerationVaries by complexityGenerating new tests with Paragon
Test RunsPer test executedRunning code tests, autotest, or performance tests
Evolving TestsPer PR analyzedProposing test updates when code changes
Agent SessionsBased on duration and modelStandard and Grind mode agent sessions
CLI UsageBased on tokens processedCommands, context queries, and code generation

More complex tasks consume more PCUs. For example, a large PR review or a multi-hour Grind session uses more than a quick single-file fix.

What Doesn't Consume PCUs

PCUs are not consumed when Paragon is:

  • Waiting for your input or approval
  • Waiting for CI/CD or test suites to complete
  • Setting up and cloning repositories
  • Idle or in sleep mode between actions

Model Tiers and PCU Usage

The model you choose affects PCU consumption:

ModelSpeedPCU Cost
FastFastestLowest
MidBalancedModerate
MaxSlowestHighest

Use Fast for simple tasks and Max for complex debugging or architecture-level work.

Plans

PlanPCU AllocationOverage
FreeIncluded monthly allowancePauses until reset
DeveloperHigher monthly allowancePurchase additional PCUs
StartupTeam-pooled PCUsPurchase additional PCUs
  • Free plan PCUs reset monthly
  • Purchased PCUs do not expire
  • Startup plans pool PCUs across all team members

Tracking Usage

Dashboard

Your PCU balance and usage are displayed on the Home page:

  • Current PCU balance
  • Active plan
  • Usage breakdown by feature (reviews, testing, agent, CLI)

Click "Show All Stats" to see detailed consumption statistics.

Per-Session Usage

Each agent session and test run shows its PCU cost, so you can see exactly how much each task consumed.

Tips to Optimize PCU Usage

  • Be specific with prompts — Clear, scoped instructions reduce unnecessary exploration
  • Choose the right model — Use Fast for simple tasks, Max only when needed
  • Split large tasks — Break complex work into focused sessions rather than one large session
  • Use Grind mode wisely — Grind uses more PCUs due to parallel workers, but saves time on large tasks
  • Review before re-running — Check results before triggering another run