Business judgment on AI products
Reads local logs from coding tools to show which task the money went to, and whether any code was actually delivered.
01
Start inside the user's day · Public facts + workflow reasoning · 2026-08-28
Reads local logs from coding tools to show which task the money went to, and whether any code was actually delivered.
Public materials do not yet show how users complete this job today or what they replace.
The product targets friction in this job, but public user evidence does not yet show the cost, frequency, or consequence of leaving it unsolved.
The trend is that AI coding became a new line of spend, so bookkeeping follows. The entry is engineering leads reconciling with finance: split the bill by project, model, and whether code shipped. Free first to earn trust; a paid hook is still unclear.
It promises a simpler way to complete this job: Reads local logs from coding tools to show which task the money went to, and whether any code was actually delivered. The exact adoption motive and repeat use are not yet verified.
① Whether a paid tier or clear pricing appears (team sync / enterprise) — where does the first; revenue beyond sponsors come from; ② Whether "cost per merged PR" and similar output-side metrics get adopted by peers — adoption; means they defined the standard; ③ Whether stars clear 15k in three month…
Keep watching. It promises a simpler way to complete this job: Reads local logs from coding tools to show which task the money went to, and whether any code was actually delivered. The exact adoption motive and repeat use are not yet verified.
when your users already produce the data nobody is reading, build the "reader of existing data" instead of the "requires instrumentation first" tool — it costs a tenth of the distribution effort. One `npx codeburn` reads the entire history; that is the first cause of its growth.
Free and open source (MIT); no paid product today. Sponsor links on GitHub; team sync; marked preview. The founder says plainly on launch: "it's already free, nothing to unlock." ① Whether a paid tier or clear pricing appears (team sync / enterprise) — where does the first; revenue beyond sponsors come from; ② Whether "cost per merged PR" and similar output-side metrics get adopted by peers — adoption; means they defined the standard; ③ Whether stars clear 15k in three month…
Reads local logs from coding tools to show which task the money went to, and whether any code was actually delivered.
It promises a simpler way to complete this job: Reads local logs from coding tools to show which task the money went to, and whether any code was actually delivered. The exact adoption motive and repeat use are not yet verified.
An English validation note will follow from the public evidence.
The product claims to help users complete: “Reads local logs from coding tools to show which task the money went to, and whether any code was ac”. User evidence has not yet verified pain intensity or the cost of doing without it.
The assessment is recorded; an English explanation is pending.
The assessment is recorded; an English explanation is pending.
The assessment is recorded; an English explanation is pending.
02
Market comparison
The Chinese–English market comparison is not complete yet. A conclusion follows only after its coverage and verifiable evidence are recorded.
03
The call and next move come first; the full read retains the evidence and counterevidence.
A free, local, open-source ledger for AI coding spend: no keys uploaded, it reads the session logs your tools already write to disk — Claude Code, Cursor, Codex, Copilot, and ~37-40 more — and breaks token and dollar spend down by task, model, project, and PR, then goes one step further and tells you whether the money actually shipped code.
Resham (ProductHunt @iamtoruk), an open-source project under the GitHub org AgentSeal.
His stated motivation: burning ~$1,400 in a week on Claude Code, with no idea where it went,
so he built it and open-sourced it. The launch thread also shows Aditya V. Singh
(@adityavsingh, self-described ex-strategy/venture, now in an operating role) answering as a
co-builder.
Read: writing the accounting tool only after burning real money is the most credible origin story in this category. And the bookkeeper's obsession — "mark it unattributed rather than smear it somewhere plausible" — is the habit of someone who has actually done financial judgment, not a growth hacker.
What it deliberately does not do: no shell-client wrapper, no cloud billing platform, no invoice-grade reconciliation. No account and no telemetry by default; data never leaves the machine.
Figuring out your AI coding spend used to mean two things: provider pages, each with its own number (Claude.ai's total, Cursor's total) that never add up, or handing API keys to a cloud aggregator and swallowing the privacy/compliance cost.
CodeBurn takes the third route — read the session files that already exist on disk. Every AI coding tool writes detailed session logs locally; nobody was reading them. It replaces the "open every vendor dashboard and stitch the bills together" ritual, and pre-answers the next question too: did any of it become a merge.
Free and open source (MIT); no paid product today. Sponsor links on GitHub; team sync marked preview. The founder says plainly on launch: "it's already free, nothing to unlock."
Read: classic "build the tool, earn trust, monetize later." The valuable bet is the "cost per useful unit of work" framing — whoever makes that metric the standard owns the category's pricing power. No evidence yet that they've decided where to charge.
| Dimension | Call |
|---|---|
| Founder-product fit | Born from a $1,400-a-week bill and first-hand pain; the conservative attribution habit comes from real financial judgment |
| Product insight | Sees two facts at once — the bill is a black box and session logs sit unread; "attribute to the PR" and "did the spend ship" are angles nobody else took |
| Execution quality | v0.9.20, five surfaces from one local source, 40 tool parsers, an MCP server; high engineering completeness |
| Timing | Right as runaway agent spend becomes a common pain; landing on the same need as Decant confirms the category exists |
One of the most serious independents in the AI-observability lane, and its direction — from "how much" to "what did it produce" — is the industry's correct next step.
The dashboard is not the asset; the attribution philosophy is: numbers that can't be traced stay "unattributed" rather than being smeared onto something plausible. Users praised this explicitly in the launch thread — for a finance tool, trust is the product. Their own data proves the rule: one unbounded directory correlation once attached 129 of 131 sessions to a single PR, and that lesson was written into the product's constraints.
Two things decide whether this moves from "worth watching" to "strong pick." First, monetization. Free plus sponsors won't sustain chasing format changes across 40 tools; team sync and the budget guard are obvious enterprise hooks but nothing is priced yet. Second, the "Optimize applies the fix itself" line — sliding from measurement tool toward an auto-optimizing agent is either the moat or a fight with the very tools it monitors, and the choice isn't made.
Competition: Decant and others sit in the same lane, and vendor-native usage pages keep improving. CodeBurn's stack of local-first, key-free, honest attribution is hard to copy quickly.
① Whether a paid tier or clear pricing appears (team sync / enterprise) — where does the first revenue beyond sponsors come from ② Whether "cost per merged PR" and similar output-side metrics get adopted by peers — adoption means they defined the standard ③ Whether stars clear 15k in three months and coverage keeps pace with new agents — coverage is the life-or-death line for this category
Product logic: when your users already produce the data nobody is reading, build the
"reader of existing data" instead of the "requires instrumentation first" tool — it costs a
tenth of the distribution effort. One npx codeburn reads the entire history; that is the
first cause of its growth.
Pricing structure: none, not disclosed — but the "free + sponsors + enterprise features in preview" staging is worth copying: let the tool grow into the de facto standard, then layer the enterprise tier.
Worth watching. Real pain, solid engineering, fast growth (9.3k stars in four months), and an attribution philosophy that is unique in the category. What's missing is a business model and a decision on the auto-optimize direction; those two determine how big it gets. Note the three checks above and revisit in three months.
05
Use these searches when the official site is missing or the current link is only a lead.