bough and the other history tools

Several tools read the history Claude Code leaves on your disk. They answer different questions, and the useful thing is knowing which question you have.

The tools below are Claude Code tools, because that is where the ecosystem is. bough also reads OpenAI Codex CLI and Pi, and if you want the same view of that work there is not much else to compare it against yet.

The short version: bough is for looking back at work that is finished. If you want to know what is running right now, or what you are spending this month across everything, or you want to read a conversation from last Tuesday through from the top, one of the others is the better tool and this page will say so.

What each one is for

The difference on cost

bough and the usage tools price the same history from the same published rates, and they agree. On two projects checked figure by figure the totals come out the same to the cent.

What differs is how finely the money can be attributed. The usage tools report by day, by month and by session, and the session is where they stop. It is an opaque hash that can cover weeks of unrelated work, so "this session cost $968" is both true and unanswerable: there is no way to ask which part of it was the feature and which was the afternoon spent stuck on one bug.

bough already works out where one piece of work ends and the next begins, so the same money lands on things you recognise. That is the whole difference, and it is what makes these answerable:

• What did building that feature cost?
• That debugging session where I went round in circles, what did it come to?
• Which day of this project was the expensive one?
• How many tokens did I burn on this project, end to end?

Neither replaces the other. A monthly total to check against a bill is a question about your account, and bough has nothing to say about it. What one task cost is a question about your work.

Where bough is the wrong choice

You want your spend by day or by month, or a figure to check against a bill. Use ccusage. It has been doing this for a long time, it is very good at it, and it reports across everything you have done at once, where bough reads one project at a time.

You want to know if you are about to run out of quota. Use Claude Code Usage Monitor, which forecasts against your actual limits. bough only looks backwards and has no opinion about what happens next.

You want to read a conversation end to end. Use Claude Code History Viewer. It searches across eleven different coding assistants, not only Claude Code, and reading a thread through is exactly what it is built for. bough can search your prompts, but it answers with rings on a diagram rather than a list of results, and it only ever shows you one prompt at a time.

You want to see your running sessions. There are a dozen good menu bar apps and status lines for that. bough does not watch anything live and will not tell you a session needs attention.

You only use a coding agent occasionally. The shape of a week with four prompts in it is not interesting. bough gets useful somewhere around a few hundred prompts on one project, when you genuinely cannot remember what happened.

Where bough is the right choice

You have lost track of a long stretch of work. Three weeks head down on one repository, and someone asks what you have been doing. A list of 181 prompts does not answer that. The shape does, in about two seconds.

You want to know what one piece of work cost. Not the month and not the session: the feature, the refactor, the day that went badly. Every task and every day carries its own figure, and the drawing sizes each piece of work by what it was charged, so the expensive parts are the ones you can see without reading anything.

You want to see the difference between work that landed and work that circled. Commits are drawn on the tasks they belong to, so a day that produced four commits looks different from a day that produced none. Everything else in the drawing is inferred; a commit either exists in your repository or it does not, and it is the one claim you can go and check.

You want to know where the hard parts were. bough marks work that took many attempts on the same files. It is a guess, it is labelled as a guess, and it is off by default. Which brings us to the honest part.

What is not settled

bough works out where one task ends and the next begins from pauses and from which files were being touched. That segmentation is fitted to one person's history, and it has one external check: if the boundaries were right, commits should cluster near the ends of tasks. Measured across four projects, they land at mean position 0.55 through a task, against 0.50 for pure chance. That is a slight lean, not the clustering you would want to see.

So the day and prompt boundaries are solid, because those come from timestamps and records rather than from a heuristic. The task boundaries in between are a best guess and may not match how you remember the work.

The struggle score sits on top of that segmentation and is unproven, which is why it is off unless you ask for it.

None of the tools above have this problem, because none of them try to infer structure. They show you what is in the file. bough is attempting something harder, and the price is that some of it is a guess.

What they have in common

Everything on this page runs locally and reads files already on your disk. Nothing here sends your history anywhere. That appears to be a settled norm in this corner of the ecosystem, and it is a good one.

Using more than one

These are not really competitors. Plenty of people will want both, because the questions sit at different sizes: what am I spending this month, and what did that one piece of work cost. They read the same files and neither writes to them, so running both is fine.