tokentoll 0.4.0

Catch LLM cost changes in code review. Infracost for LLM spend.

tokentoll

Catch LLM cost changes in code review. Infracost for LLM spend.

A CLI tool and GitHub Action that statically analyzes your code for LLM API calls, estimates their cost, and shows you the cost impact of every change -- in your terminal or as a PR comment. Zero runtime dependencies.

The Problem

A single model swap from gpt-4o-mini to gpt-4o increases costs 15x. A new API call in a hot path can add $10,000/month to your bill. These changes hide in normal code review.

tokentoll finds LLM API calls in your code, estimates their cost, and shows you the cost impact of every change -- before it hits production.

Quick Start

pip install tokentoll

# Scan current directory for LLM API calls and their costs
tokentoll scan .

# Show cost impact of your last commit
tokentoll diff HEAD~1

# Compare two branches
tokentoll diff main..feature-branch

GitHub Action

name: LLM Cost Diff
on:
  pull_request:
    paths:
      - "**.py"

permissions:
  pull-requests: write

jobs:
  cost-diff:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0

      - uses: Jwrede/tokentoll@v1
        with:
          calls-per-month: "5000"

What It Detects

SDK	Patterns	Status
OpenAI	chat.completions.create, responses.create	Supported
Anthropic	messages.create, messages.stream	Supported
Google GenAI	models.generate_content	Supported
LiteLLM	completion, acompletion	Supported
LangChain	ChatOpenAI, ChatAnthropic, init_chat_model	Supported
JS/TS SDKs	--	Planned

Example Output

tokentoll scan

LLM API Calls Detected
============================================================

File: src/agents/summarizer.py
  Line 42: openai client.chat.completions.create
           Model: gpt-4o | Max tokens: 4096
           Est. cost/call: $0.03 | Monthly (1000 calls/month per call site): $26.50

  Line 78: openai client.chat.completions.create
           Model: gpt-4o-mini | Max tokens: 1000
           Est. cost/call: $0.000301 | Monthly (1000 calls/month per call site): $0.30

--
Total estimated monthly cost: $26.80
  1000 calls/month per call site

tokentoll diff

LLM Cost Diff: main..feature-branch
============================================================

+ ADDED    src/agents/rewriter.py:35
           openai | Model: gpt-4o
           Est. cost/call: $0.03 | Monthly: +$26.50

~ MODIFIED src/agents/summarizer.py:42
           openai | Model: gpt-4o -> gpt-4o-mini
           Est. cost/call: $0.03 -> $0.000301 | Monthly: -$26.20

--
Monthly cost impact: +$0.30
  Added: 1 | Changed: 1 | Removed: 0
  1000 calls/month per call site

How It Works

  Source Code (.py files)
         |
         v
  +-------------+     +------------------+
  | AST Scanner |---->| SDK Detectors    |
  | (ast.parse) |     | OpenAI, Anthropic|
  +-------------+     | Google, LiteLLM  |
                       | LangChain        |
                       +------------------+
                              |
                              v
                       +------------------+
                       | Pricing Engine   |
                       | 2200+ models     |
                       | Auto-cached      |
                       +------------------+
                              |
                  +-----------+-----------+
                  |                       |
                  v                       v
           +------------+         +-------------+
           | Scan Report|         | Diff Engine  |
           | (costs)    |         | (old vs new) |
           +------------+         +-------------+
                  |                       |
                  v                       v
           +------------+         +-------------+
           | Table/JSON |         | Table/JSON/  |
           |            |         | PR Comment   |
           +------------+         +-------------+

1. Parses Python files using the ast module to find LLM API calls
2. Multi-pass constant propagation resolves model names through variables, os.getenv() fallbacks, class attributes, constructor args, dict contents, and **kwargs unpacking
3. Looks up pricing from a local cache (sourced from LiteLLM, 2200+ models)
4. For diff mode: compares calls between two git refs and computes the cost delta
5. Outputs a cost report as a table, JSON, or GitHub PR comment

CLI Reference

tokentoll scan [PATH...] [--format table|json|markdown] [--calls-per-month N]
tokentoll diff [REF] [--base REF] [--head REF] [--format table|json|markdown|github-comment]
tokentoll update    # Update bundled pricing data

Pricing Data

Pricing is bundled and works offline. To update to the latest prices:

tokentoll update

Pricing data is sourced from LiteLLM's model_prices_and_context_window.json and covers 300+ models across OpenAI, Anthropic, Google, AWS Bedrock, Azure, and more.

Smart Variable Resolution

Real codebases rarely pass model names as string literals. tokentoll's multi-pass constant propagation engine follows:

DEFAULT_MODEL = os.getenv("MODEL", "gpt-4o")

class Config:
    model: str = DEFAULT_MODEL

config = Config()
kwargs = {"model": config.model, "max_tokens": 2000}
client.chat.completions.create(**kwargs)
# tokentoll resolves: model="gpt-4o", max_tokens=2000

• Variable assignments (MODEL = "gpt-4o")
• os.getenv() / os.environ.get() fallback values
• Function default parameters
• Class attribute defaults
• Constructor argument propagation
• Dict literal and subscript contents
• **kwargs unpacking

Limitations

• Cannot resolve models loaded from external config files or databases at runtime (these are flagged as dynamic but not priced)
• Token estimates use a character/4 heuristic unless tiktoken is installed
• Monthly estimates assume uniform call volume (configurable via --calls-per-month)
• Python only for now (JS/TS support planned)

License

MIT