PermitGraph Docs

Cost Controls

How model usage stays bounded during Deep Agent review.

What this is

Cost controls track and limit live model usage.

When to use it

Use this page before running OpenRouter-backed investigations or live validation.

Command or example

export OPENROUTER_API_KEY=<key>
uv run --extra deep-agent agent-permit investigate .agent-permit/runs/<run_id>

Output to expect

Usage artifacts record token counts, cached tokens, cache-hit ratio, model calls, and run metadata.

How to interpret it

Prompt caching and bounded artifacts reduce model cost. The scanner remains deterministic and local-first.

Common mistakes

  • Running live validation without checking API key and spend limits.
  • Asking the model to read the whole repository instead of bounded artifacts.
  • Ignoring cache metrics after a demo run.

On this page