About

Built to make prompt token usage visible, debuggable and cheap.

Why it exists

When a prompt contains tools, RAG chunks and a long chat history, the token bill becomes a black box. prompt-flamegraph turns that black box into a picture you can reason about.

Design principles

No required dependencies
No server, no telemetry, no proxy
One function call, one output file
Pluggable tokenizers
Open source under LGPL-3.0-or-later

When to use it

Your prompts keep growing and you do not know why
You want to compare two prompt versions
You suspect duplicate context or too many tools
You need to explain token cost to a teammate

Author & license

prompt-flamegraph is maintained by Fjjjuv and released under the GNU Lesser General Public License v3.0 or later.

GitHub

@fjjjuv

github.com/fjjjuv

Dev.to article

The story behind the tool, with step-by-step examples.

Read on Dev.to

PyPI

Install the latest release.

pypi.org/project/prompt-flamegraph

Changelog

Latest changes from the repository.

Version	Highlights
`0.4.1`	Windows console and redirected-stream robustness, BOM-tolerant JSON input, CI matrix and trusted PyPI publishing.
`0.4.0`	Dark mode: theme="auto\|light\|dark" and --theme CLI flag, CSS custom properties, persisted sun/moon toggle.
`0.3.3`	Hardening release: token-accuracy fixes, error handling, security.
`0.3.2`	OpenAI Responses API payloads, top-level exports, Markdown waste report, accessible tooltips, token-accuracy and CLI fixes.
`0.3.1`	README redesign, PyPI/GitHub version alignment.
`0.3.0`	API payload adapters, model presets and dynamic pricing, CI budget gate, framework adapters, JSON export, aggregation buckets, LGPL-3.0-or-later license.
`0.2.4`	Website link in README and PyPI metadata.
`0.2.3`	Version sync, README updates with PyPI and Dev.to links.
`0.2.2`	README badges, waste detection example, copyright year updates.
`0.2.1`	Demo screenshot in README.
`0.2.0`	Initial release with HTML/SVG/Markdown flamegraphs, waste detection, diff, terminal output and CLI.