🛡 AgentGuard

Pre-execution cost estimator for agentic AI systems

AgentGuard analyzes your multi-file agentic codebase and estimates the minimum and maximum possible LLM API cost — without running any code or calling any APIs.

Install

pip install -e .

Usage

agentguard --root_dir ./my_agent
agentguard --root_dir ./my_agent --model claude-3-5-sonnet
agentguard --root_dir ./my_agent --input_tokens 400 --output_max 1000
agentguard --root_dir ./my_agent --json_out report.json
agentguard --root_dir ./my_agent --quiet --dag

What It Detects

• API calls inside loops (cost multiplies with iteration count)
• Nested loops (exponential scaling)
• Missing loop depth limiters (unbounded execution)
• Repeated prompts (caching opportunity)
• Large static prompts (high baseline cost)
• High-complexity tasks (reasoning tax ×20)

Supported Models

gpt-4o, gpt-4o-mini, gpt-4-turbo, claude-3-5-sonnet, claude-3-haiku, claude-3-opus, gemini-1.5-pro, gemini-1.5-flash

Architecture

CLI → Crawler → Analyzer (AST/regex) → Simulator (math model) → Reporter