Billing & Credits

Critique uses a credit-based billing system to manage access to its AI models, separating automated background operations (PR reviews, Remedy) from interactive workspace chat.

1. Credit Allocations

Credits represent a unified currency for AI compute across the platform. They are allocated based on your organization's subscription plan tier (Standard vs. Ultra).

• Standard Plan: Provides a monthly pool of credits sufficient for teams using the standard suite of models (e.g., Gemini Flash, Claude Sonnet, Qwen).
• Ultra Plan: Grants a larger monthly pool and unlocks access to high-capability reasoning models (e.g., Claude Opus, GPT-Pro series) which inherently consume more compute per token.

Base credits reset at the beginning of each billing cycle. If your organization exhausts its allocation before the cycle ends, reviews may be temporarily paused unless overage billing is enabled.

2. PR Reviews vs. Chat

Critique separates the cost tracking of its two main product surfaces:

• Automated PR Reviews: Running the full review pipeline (Scout, specialists, lead synthesis) is a backend operation. The cost is calculated based on the precise number of tokens consumed by the specific runtime models selected in your repository policies.
• Remedy Runs: Starting an autonomous remediation agent also consumes credits based on the execution model's token usage and the number of loops required to pass validation.
• Workspace Chat: Interactive chat sessions use a separate pool of chat-optimized models. While chat interactions do consume compute, they are generally less intensive than full-repository PR reviews. The dashboard tracks chat usage distinctly from automated pipeline usage.

3. The Credit Floor

To prevent unexpectedly draining an organization's remaining credit balance on a single large pull request, Critique implements a creditFloor mechanism for its models.

Each model in the catalog (managed in lib/models/catalog.ts) is assigned a specific creditFloor value. For example, high-end models like claude-opus have a significantly higher floor (e.g., 37 credits) than standard models.

If your organization's remaining credit balance falls below a model's creditFloor, Critique will automatically prevent that model from being selected for a review run. The system will gracefully fall back to a more efficient standard model (if permitted by policy) or pause the review until credits are replenished, ensuring you never accidentally incur massive overage charges for a massive diff.