CodeBurn Review (2026): Pricing, Features & Honest Verdict

Reviewed by MakerStack · Published · 8 min read

CodeBurn is a free, open-source tracker that reads the session files Claude Code, Cursor, Codex and 37 other agents already write to your disk, then turns them into an itemised bill by model, project and pull request. Best for: developers with no idea where their AI coding spend goes. Price: free, MIT licensed. Rating: 8/10.

What is CodeBurn?

CodeBurn answers a question that did not exist two years ago and now costs teams real money: where did all that go? Every AI coding tool writes session files to your machine as you work. CodeBurn reads them, across 40 different tools, and turns the raw logs into a breakdown of spend by model, by project, by task and by the pull request the work ended up in. The tagline on the homepage is “Your AI Bill, Itemized”, and that is precisely what it does.

Nothing is uploaded. There is no account, no sign-up and no server. You run npx codeburn and it parses files that are already sitting on your disk. It ships as a menu bar app, a desktop app and a terminal tool, across macOS, Windows and Linux. It is MIT licensed with the source on GitHub, where it has gathered roughly 9,250 stars and 728 forks since the repository was created in April 2026, and it was last updated the day we tested it. The maker organisation is Agentseal, a security company that works on red-teaming AI agents and scanning MCP servers, with Resham Joshi and Aditya Vikram Singh credited on the launch.

What are CodeBurn’s key features?

Reading session files instead of proxying traffic

This is the architectural decision everything else follows from. Most cost-tracking tools want to sit between you and the model as a proxy, which means routing your code through someone else’s infrastructure and trusting them with it. CodeBurn reads the artefacts your tools already produce locally. The privacy benefit is obvious. The subtler benefit is coverage: because it parses files rather than intercepting API calls, it can cover 40 tools without integrating with any of them, including ones that would never let a proxy in. The subtler cost is fragility, which we come back to below.

Attribution to commits and pull requests

The differentiated feature. CodeBurn matches spend to the commits and pull requests the work actually produced, so you can distinguish shipped work from abandoned exploration. Knowing you spent $340 last month is mildly interesting. Knowing that $210 of it landed in merged pull requests and $130 went to sessions that produced nothing is the number that changes behaviour. No other tool in this category we have looked at closes that loop.

Waste detection and an A-to-F grade

Beyond measurement, CodeBurn identifies specific waste patterns: duplicate file reads, context bloat and what it calls ghost agents, which are sessions still burning tokens that nobody is watching. It then grades your whole setup from A to F on an efficiency score built from one-shot rate, cache hit rate and retry tax. The boldest claim on the page is that fixes are applied for you, with an undo available for each. That is a much stronger promise than flagging problems, and it is the claim we would most want to verify hands-on before relying on it.

Live plan-limit tracking in the menu bar

The menu bar app keeps the day’s spend next to your clock and tracks plan usage live for each agent, showing how much time remains in each rate-limit window. For anyone on a subscription with usage windows rather than pure pay-as-you-go, this is the practical feature: it turns “why did my agent stop responding” into a number you can see before it happens. On Linux the same figure lives in the GNOME top bar.

Cost flow from model to project

The desktop app traces which model spent into which repository over any date range you choose, projects a month-end total and compares your pace against last month. For a team lead trying to work out whether the expensive model is actually earning its premium on a specific codebase, this is the view that answers it. Spend split by activity type, separating coding from debugging from feature work with turn counts and one-shot rates, is the same idea at finer grain.

How much does CodeBurn cost?

Nothing. CodeBurn is free and MIT licensed, with no paid tier, no premium version, no hosted plan and no usage limits. There is no pricing page because there is nothing to price.

That is unusual enough in this category to be worth stating plainly. The obvious commercial move for a tool that watches your AI spend is a team dashboard with seats and a monthly fee, and CodeBurn has not made it. The likely reason is visible in the parent organisation: Agentseal sells AI agent security, and CodeBurn is a well-executed open-source project that puts the company in front of exactly the developers Agentseal wants to reach. That is a legitimate model and a common one, but it means the tool’s future depends on a commercial strategy it is not itself part of.

What does “free and open source” actually commit to here?

Less than people assume, and the distinction matters when you are deciding whether to build a workflow on top of it.

What you actually get: an MIT licence, which is about as permissive as licences come, and source code you can fork if the project is abandoned. Roughly 9,250 stars and 728 forks mean a real community rather than a weekend project, and the repository was pushed to on the day we looked, which is the strongest signal of all. Nothing runs on a server that can be switched off, so a shutdown does not break your installation.

What you do not get: any support commitment, any uptime guarantee, and any assurance the project continues. There are 41 open issues. Two named makers are behind it, working at a security company whose main product is something else. If a session file format changes and nobody has time to fix the parser that week, your numbers go wrong quietly rather than loudly.

That last risk is the structural one. CodeBurn’s coverage of 40 tools is its best feature and its biggest maintenance liability, because it depends on 40 undocumented file formats it does not control. Any of those vendors can change their session format in a routine release, and CodeBurn finds out when a user reports it. This is not a reason to avoid the tool. It is a reason to sanity-check its totals against your provider’s own billing occasionally rather than treating it as an authoritative ledger.

How much of the marketing can you verify?

Most of it, which is more than we can usually say, with two small discrepancies worth noting.

The tool count is inconsistent across sources. The website says “Supported tools, 40 and counting” and describes “Claude Code, Cursor, Codex, or 37 others”, which is internally consistent at 40. The GitHub repository description still says 31 tools. The site is almost certainly current and the repository blurb stale, but it is the sort of drift worth noticing on a project whose entire value is accurate counting. The star count on the homepage was also lagging the actual figure on GitHub when we checked.

The Product Hunt listing claims over 150,000 developers use it. We could not find that figure anywhere on the vendor’s own site and we are not repeating it as fact. The number we can verify is roughly 9,250 GitHub stars, which is strong for a project four months old and is the honest way to describe its traction.

Everything else checks out. The MIT licence is confirmed via the GitHub API rather than inferred from a badge. The named tools are listed explicitly. The local-only architecture is consistent with a tool distributed as npx codeburn with no account step. And the “last updated today” activity is real.

PlanPricePlan FeaturesBest For
Free$0Every feature, 40 supported tools, unlimited usage, MIT licensed, no account requiredEveryone; there is no paid tier and no hosted plan

Who is CodeBurn best for?

Use CodeBurn if you or your team spend meaningfully on AI coding tools and cannot currently attribute that spend to anything. That is most teams right now. It is especially useful if you run several agents at once, because the per-tool dashboards each vendor provides cannot see each other and CodeBurn produces one ledger across all of them. It is also the obvious choice if you work somewhere that would never approve routing source code through a third-party proxy, since nothing leaves the machine and there is no vendor to put through procurement.

Skip CodeBurn if you use exactly one AI coding tool and its native dashboard already tells you what you need, because the marginal value of a unified view of one thing is small. Skip it too if you need a supported product with an SLA and someone to call, which a free two-person open-source project is not. And do not adopt it as your billing system of record: use it to find waste and understand patterns, and reconcile against your provider’s invoices for anything that matters financially.

Best CodeBurn alternatives

ccusage is the tool CodeBurn positions itself against by name in its own meta description. Free, open source and terminal-first, focused primarily on Claude Code usage. Lighter and simpler, with nothing like CodeBurn’s 40-tool coverage, commit attribution or graphical apps. Try it first if all you want is a quick token count for one tool.

Your provider’s own dashboard, whether the Anthropic Console, Cursor’s usage page or GitHub Copilot’s admin reporting, is free, authoritative for billing and requires no installation. It is the right answer when you use one tool. It cannot combine spend across vendors, attribute cost to a pull request, or tell you what fraction of your spend produced merged code.

Langfuse or Helicone are the serious observability platforms for teams building LLM applications rather than just coding with them. Both have free tiers and paid plans, both require instrumenting your calls, and both go far deeper on traces and evaluation. Overkill for tracking your own coding spend, correct if you are shipping an AI product and need production observability.

Final verdict: is CodeBurn worth it?

Yes, and the decision is easy because it costs nothing and runs entirely on your machine. The problem it solves is real and getting worse: AI coding spend is now a meaningful line item for a lot of teams, it is spread across several vendors who each show you only their own slice, and almost nobody can say which of it produced shipped code. CodeBurn answers that in one install, with no account and no data leaving your laptop. The commit and pull-request attribution is the part that turns it from a curiosity into something that changes how you work, and we have not seen another tool do it.

The caveats are about durability rather than quality. There is no paid tier, no support commitment and no SLA, and the project depends on parsing 40 file formats controlled by other companies, any of which can change without warning. Two people maintain it alongside jobs at a security company that sells something else. Treat CodeBurn as an excellent free diagnostic rather than infrastructure: install it, find out where your money is actually going, act on what it shows you, and check its totals against a real invoice before you make a budgeting decision on them.

CodeBurn Pros & Cons

What We Like

  • Completely free and MIT licensed, with around 9,250 GitHub stars and commits landing daily
  • Runs entirely on your machine with no account, no upload and no proxy sitting in front of your source code
  • Covers 40 AI coding tools in a single ledger, which no individual vendor dashboard can do
  • Attributes spend to the commits and pull requests it produced, separating shipped work from abandoned exploration
  • Ships as a menu bar app, a desktop app and a terminal tool across macOS, Windows and Linux

What Could Be Better

  • No paid tier means no support commitment, no SLA and no guarantee the project continues
  • Depends on parsing 40 undocumented session file formats it does not control, any of which can change in a routine release
  • Maintained by two people alongside jobs at a security company whose main product is something else
  • The GitHub repository still advertises 31 supported tools where the website says 40, and the site's star count lags the real figure

CodeBurn FAQ

What is CodeBurn?

CodeBurn is a free, open-source tool that tracks AI coding spend. It reads the session files that Claude Code, Cursor, Codex, Copilot, Gemini and 35 other agents already write to your disk, then breaks the cost down by model, project, task and the pull request the work produced. Nothing is uploaded and there is no account. You run it with npx codeburn.

How much does CodeBurn cost?

Nothing. CodeBurn is free and MIT licensed with no paid tier, no premium version, no hosted plan and no usage limits. There is no pricing page because there is nothing to price. The maker organisation, Agentseal, sells AI agent security separately, which is the likely commercial reason the tool itself is given away.

Is CodeBurn worth it?

Yes, and the decision is easy because it costs nothing and runs entirely locally. AI coding spend is now a real line item spread across several vendors who each show you only their own slice. CodeBurn unifies it in one install. Treat it as an excellent free diagnostic rather than infrastructure, and reconcile its totals against a real invoice before making a budgeting decision.

What are the best CodeBurn alternatives?

ccusage is the tool CodeBurn names itself against: free, open source and terminal-first, focused mainly on Claude Code. Your provider's own dashboard, such as the Anthropic Console or Cursor's usage page, is authoritative for billing but cannot combine vendors. Langfuse and Helicone are the serious observability platforms if you are shipping an AI product rather than just coding with one.

Does CodeBurn offer a free plan?

CodeBurn is entirely free. There is no free tier as such because there is no paid tier to contrast it with: the whole tool is MIT licensed and every feature is available to everyone. What you do not get is any support commitment, uptime guarantee or assurance the project continues, which is the honest trade for paying nothing.

Who is CodeBurn best for?

Developers and teams spending meaningfully on AI coding tools who cannot currently attribute that spend to anything, especially those running several agents at once since per-vendor dashboards cannot see each other. It also suits organisations that would never approve routing source code through a third-party proxy, since nothing leaves the machine and there is no vendor to put through procurement.

Disclosure: MakerStack is funded by featured placement fees, sponsor slots and a small number of affiliate links. Nobody paid for this review. Where any of those does apply to a review, we say so on the page. The scoring criteria are the same in every case. See our editorial policy.

This review describes CodeBurn as we tested it on August 12, 2026. Pricing, plans and features may have changed since. We don't update reviews on request; see our update policy.