Cost Management

AI-powered features consume LLM tokens. Use these controls to keep costs predictable.

Free-Tier Commands

These commands run without any LLM calls or API keys:

Command	What It Does
impact	Deterministic impact analysis from git diff
plan	Coverage gap detection and test recommendations
train --no-enrich	Build route-families manifest (scanner only)
feedback	Ingest recommendation outcomes
cost-report	View past LLM cost breakdown
traceability-capture	Extract test-file relationships
traceability-ingest	Merge traceability mappings

Budget Flag

Cap LLM spending per run with --budget-usd:

npx impact-gate crew --workflow design-only --budget-usd 2.00 --path .

When the budget is exhausted, further LLM calls are rejected and the run completes with whatever results it has so far. The default budget for train is $0.50 with a max of $10.

Cost Report

View a breakdown of LLM costs from past runs:

npx impact-gate cost-report --path .

This reads the cost tracking data stored in .e2e-ai-agents/ and reports per-agent and per-model spend.

Model Routing

The model router sends different task types to different models to optimize cost:

• Simple tasks (classification, short answers) go to cheaper, faster models
• Complex tasks (test generation, code analysis) go to more capable models

This happens automatically when using the default provider configuration.

Degraded Mode

For zero-cost operation that skips all AI calls:

npx impact-gate crew --workflow quick-check --degraded-mode --path .

In degraded mode, agents that require LLM calls are skipped. Deterministic analysis (impact mapping, gap detection) still runs, so you get partial results at no cost.

Cost by Workflow

Typical costs per crew workflow run:

Workflow	Typical Cost	Range
quick-check	~$0.10	$0.05-0.20
design-only	~$1.00	$0.50-2.00
full-qa	~$3.00	$2.00-5.00