See where your LLM prompt tokens go

Lightweight, zero-dependency Python package to profile prompts with interactive flamegraphs, token-waste detection, prompt diffs and HTML/SVG/Markdown/terminal exports.

pip install prompt-flamegraph
View on GitHub Get on PyPI
Python 3.10+ LGPL-3.0-or-later Zero dependencies

Interactive flamegraph demo

Hover over the bars to see how tokens are distributed across system prompt, tools, RAG context and chat history.

Total tokens: 102 · Estimated cost: $0.000153 at 1.5e-6/token

Why use prompt-flamegraph?

A focused tool for a real problem: prompt tokens are expensive, opaque and easy to waste.

Visual token breakdown

Turn a structured prompt into an interactive HTML flamegraph and spot which category eats the budget.

Waste detection

Find duplicate text, oversized RAG context, long chat history and too many tools before calling the API.

Prompt diffs

Compare two prompt versions and visualize added, removed and changed tokens.

Multiple exports

Export to HTML, SVG, Markdown, JSON or a colored terminal bar chart.

Zero dependencies

Pure Python, no server, no dashboard, no telemetry. Optional tiktoken and rich extras.

CLI & API

One function call in Python or a single command in your terminal.

Explore all features

Quick start

Install it, call one function, open the HTML report.

python 
from prompt_flamegraph import profile_prompt

prompt = {
    "system_prompt": "You are a helpful coding assistant.",
    "tools": ["..."],
    "rag_context": {"doc_1": "..."},
    "chat_history": ["..."],
}

profile_prompt(prompt, output="context.html")
See more examples

CLI examples

$ prompt-flamegraph prompt.json -o context.html --cost 1.5e-6
$ prompt-flamegraph prompt.json --terminal
$ prompt-flamegraph v1.json --diff v2.json -o diff.html
$ prompt-flamegraph prompt.json --model gpt-4o
$ prompt-flamegraph prompt.json --budget 100000
$ prompt-flamegraph --demo --cost 1.5e-6

Detect waste before paying

Get a WasteReport with concrete findings before sending the prompt to an API.

output
Wasted: 26 / 88 tokens (29.5%)
- duplicate: 3x duplicate text ('def helper():     return 'value' ') — keep only one
- duplicate: 5x duplicate text ('Hi!') — keep only one
- too_many_tools: 6 tools defined — only declare the ones the model actually calls

What the report looks like