burntop
burntop is a free, open-source tracker for AI coding assistants. A command-line sync collects token counts and estimated costs from Claude Code, Cursor and others, then feeds real-time dashboards, community leaderboards and shareable README badges.
What is burntop?
burntop is a usage tracker for developers who work with AI coding assistants. It answers a question those tools rarely answer themselves: how many tokens went through them this week, what that probably cost, and how it compares with everyone else doing the same job. The project calls itself a gamified AI usage tracking platform, and the gamification is not decoration. Leaderboards, streaks, levels and achievements are the main interface.
Collection happens locally. A command-line client, run through bunx or installed globally with bun, npm or pnpm, walks the directories the assistants already write to: Claude Code session logs, the Cursor SQLite database, Gemini CLI chat files, and the Aider and Continue session folders. Nothing needs configuring. What leaves the machine is narrow, since the documented sync payload carries a timestamp, a model name, a source, input, output and cache token counts and a request identifier used for deduplication. Prompts and source code are absent from that schema. Syncing is rate limited to ten calls a minute per account, and signing in is GitHub OAuth only.
The web side turns those records into five things. A real-time dashboard shows totals, estimated cost, current streak and a seven-day trend. Detailed analytics break usage down by model and by tool and show when you are most productive. A community leaderboard runs in seven flavours, from total tokens and model diversity to cache efficiency, streak length, week-over-week risers, reasoning tokens and a friends-only board, filterable by period, tool, model and region. Shareable stats generate SVG badges for a GitHub README in six variants and four styles, plus Open Graph cards for social posts. A yearly Wrapped compiles top model, top tool, active days, longest streak and a projection for the year ahead, while an insights endpoint places you in percentiles against community benchmarks.
The whole thing is open source under the MIT licence, with a documented REST API and a self-hosting guide covering PostgreSQL, Docker Compose and environment setup. It is also very young: the domain was registered in January 2026 and the public repository still shows a handful of stars and fewer than thirty commits.
What it does
- Collect AI coding assistant usage automatically from local files through a command-line client
- Aggregate input, output and cache tokens into a running total with an estimated cost
- Display a real-time dashboard of tokens, spend, current streak and seven-day trend
- Rank you on a community leaderboard filtered by period, tool, model or region
- Award levels, XP, streaks and achievements as your usage accumulates
- Generate embeddable SVG badges and shareable Open Graph cards from your stats
- Produce a yearly Wrapped review with top model, top tool and next-year projections
When to use burntop / When not to
A quick filter to help you decide if burntop is the right fit.
When to use burntop
- Developers who run Claude Code, Cursor, Gemini CLI, Aider or Continue every working day
- Freelance developers who need to put an actual number on their monthly AI spend
- Technical founders and startup CTOs curious about their own token consumption
- MLOps and LLMOps engineers interested in per-model breakdowns and cache-hit efficiency
- Developers who enjoy displaying a live stats badge in their GitHub README
When not to use burntop
- Teams that need billing, client re-invoicing or shared budget controls, none of which exist here
- Organisations that require published terms, a privacy policy or a DPA before adopting any tool
- Anyone tracking non-coding AI usage such as image, video or voice generation
- Developers without a GitHub account, since GitHub OAuth is the only way to sign in
- People unwilling to install the Bun runtime that the command-line client depends on
How to use burntop
A typical end-to-end flow, from setup to results.
- Open the burntop website and click Get Started
- Sign in with Continue with GitHub, the only authentication method offered
- Install the Bun runtime, which the command-line client requires
- Run the client without installing anything: bunx burntop sync
- Or install it globally with bun add -g burntop, npm install -g burntop or pnpm add -g burntop
- Authenticate the client with bunx burntop login
- Push your local usage to the cloud with bunx burntop sync
- Review your figures from the terminal with bunx burntop stats, or on the web dashboard
- Check your standing with bunx burntop leaderboard, or on the leaderboard page
- Set your profile visibility, embed a badge in your README, and sync regularly to keep the streak alive
Pros & Cons
Pros
- Genuinely free, with no paid tier at all, no card required and no announced limits
- Open source under the MIT licence, so the code that reads your local files can be audited
- Self-hosting is documented, with Docker Compose, a deployment guide and security guidance
- One command gets you started, with no package to install first, no API key and no config file
- Five coding assistants are detected automatically, with no paths to declare
- The documented sync payload carries token counts only, not prompts or source code
- A public REST API, server-rendered badges and OG cards, with explicit HTTP caching
Cons
- No terms, no privacy policy and no legal notice: the operator is identified by no document
- No company name, no postal address and no support or legal contact, only a security address
- Nothing at all on the GDPR, retention, deletion, subprocessors or a Data Processing Agreement
- Very young and barely adopted: first archived in January 2026, a few stars and no forks
- Coverage is narrower than advertised, with five documented parsers against a claimed ten-plus tools
- Costs are estimates derived from token counts, not amounts actually billed by the vendors
- GitHub OAuth is the only way in, and the client adds a Bun runtime dependency
Pricing & Plans
A permanent free plan is the only offering. burntop is presented as Free Forever at $0, with no credit card required and no hidden fees. There is no paid tier, and consequently no entry price point in any currency.
- unlimited tracking
- all features included (leaderboards
- achievements and analytics)
- 10+ AI tools supported
- shareable stats and badges
- Yearly Wrapped
- no limits or restrictions
Data, GDPR & hosting
A consolidated view of how burntop handles your data.
GDPR overview
There is no GDPR mention of any kind. A search across every page collected, from the homepage and the login screen to robots.txt and the entire public repository, returns not a single occurrence of GDPR or CCPA; the only hits on retention concern database backups in the self-hosting guide. No privacy policy, no cookie notice, no consent banner, no named controller, no data protection officer, no Article 27 representative, no Data Processing Agreement and no subprocessor list. The only document carrying any legal weight is the MIT licence covering the source code. A responsible-disclosure procedure does exist, with a dedicated security address, but it deals with vulnerabilities rather than personal data. Nothing states that the service is non-compliant; it simply says nothing at all, which is the first point any compliance team would raise.
Who owns the data?
No terms of service and no privacy policy exist anywhere: the /privacy and /terms addresses both land on a user-profile route, and the public repository contains neither document. Ownership of the records you sync is therefore defined by no contract at all. What the technical documentation does show is narrow. The sync endpoint accepts usage metadata only, a timestamp, a model name, a source, input, output and cache token counts and a request identifier, with no prompt and no source code in that schema. Visibility is the single lever you hold: a profile flag switches between public, where everything is on show, and private, where non-followers see only your username, name and avatar. The MIT licence covers the code, not your data.
Reuse rights
Nothing grants or refuses you a right of reuse in writing, because no terms exist. In practice the platform hands your own figures straight back through its API: stats, level, achievements, dashboard breakdowns and the yearly Wrapped payload. For public profiles two endpoints require no authentication at all, an SVG badge and an Open Graph card, which is exactly what makes a README badge work, in six variants, four styles and a light or dark theme. A real-time stream is exposed over Server-Sent Events. The code itself is MIT licensed, so the whole platform may be forked, modified and self-hosted, and a deployment guide with a Docker Compose file is provided for that.
Data retention & training
Hosting summary
The vendor publishes no hosting information at all: no country, no region, no provider, no processing jurisdiction, and no mention of international transfers or standard contractual clauses. What can be observed is a network fact rather than a statement. On 30 August 2026 the domain resolved to 168.119.140.28, an address geolocated to Falkenstein, Germany, on Hetzner Online GmbH's network. That is where the site answered from on that day; it is not a commitment by the operator, and it says nothing about where user records are stored or backed up. The repository documents PostgreSQL as the database and describes several ways of deploying it, including Neon, Supabase, Railway and a local Docker container, but those pages are instructions for anyone running their own instance rather than a description of the hosted service. Self-hosting is in fact the only way to know the answer, since an instance you deploy puts the data in whatever jurisdiction you choose.
Things to keep in mind
Risks and trade-offs to weigh before adopting burntop.
- Streaks, levels and rankings reward volume, which can quietly nudge you into spending more on tokens
- A public leaderboard measures tokens burned, a figure that says nothing about the quality of the work
- Public profiles, badges and social cards expose working habits by name: rhythm, productive hours, tools used
- No contract of any kind covers the service, so nothing commits the operator on retention or deletion
- The absence of any GDPR or DPA material makes professional use inside the EU hard to justify
- The client reads sensitive local directories, including assistant session logs, on trust in the source code alone
- Depending on a very young single-maintainer project with no identified company means continuity is not assured
Setup & Integrations
Technical difficulty
Low for normal use, provided you are comfortable in a terminal. Signing in takes a single GitHub click, and one command, bunx burntop sync, does the rest: there is no API key to create and no path to configure, since the assistants' directories are detected automatically. The only prerequisite is the Bun runtime. Self-hosting is another matter, calling for Node 20+, Bun 1.3+, Docker and Docker Compose, a PostgreSQL database, Alembic migrations, a GitHub OAuth application and a 32-character authentication secret.
Deployment
Integrations
Behind burntop
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
How much does burntop cost?
Which AI tools does it track?
How do I get started?
What data actually leaves my machine?
Are my statistics public?
Is there a privacy policy or terms of service?
Can I self-host it?
Is there an API?
Is there a mobile app?
Should you pick burntop?
burntop does one thing, for a narrow audience, and does it well: it turns the token counts your coding assistants already write to disk into a dashboard, a leaderboard position and a badge you can drop into a README. For a developer who runs Claude Code or Cursor all day and has never seen the aggregate, that is genuinely useful information, and it costs nothing to obtain. The single plan is free forever, the code is MIT licensed and auditable, and one command starts the sync.
The reservations are not about the product. They are about everything around it. There is no privacy policy, no terms of service, no legal notice, no company name, no postal address and no support contact. Nothing addresses the GDPR, data retention, deletion or subprocessors. A tool whose client reads your local assistant session directories is precisely the kind of tool for which those documents matter, and here the only assurance on offer is the source code itself. That is a real assurance for a developer able to read it, and none at all for a compliance team.
The project is also very young. The domain was registered in January 2026, the first archived capture dates from the same month, and the public repository shows a handful of stars, fewer than thirty commits and no forks. Continuity is guaranteed by nothing.
The trade is explicit, and honest in its way: nothing to pay, and nothing contractual either. For individual developers curious about their own AI consumption, and for anyone comfortable running the stack themselves, burntop is worth the five minutes it takes to try. For regulated environments or team-wide reporting, it is not a candidate yet.
- Choosing a selection results in a full page refresh.
- Opens in a new window.