Business judgment on AI products
Turns local coding-assistant sessions into a searchable ledger of spend, files touched, and what the assistant actually did.
01
Start inside the user's day · Public facts + workflow reasoning · 2026-08-28
Turns local coding-assistant sessions into a searchable ledger of spend, files touched, and what the assistant actually did.
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 teams wanting a replay of what the AI did and where the money went. The entry is a local session audit: see cost for free, then sell the missing knowledge. Overlap with spend trackers means the pain is already common.
It promises a simpler way to complete this job: Turns local coding-assistant sessions into a searchable ledger of spend, files touched, and what the assistant actually did. The exact adoption motive and repeat use are not yet verified.
① Whether stars clear 1,000 within three months and the HN attention converts to sustained; adoption — current public heat is low (10 points / 0 comments); ② Whether any user publicly reports that distilled scripts are actually being reused — it is; the only feature that differentiates it from a led…
Keep watching. It promises a simpler way to complete this job: Turns local coding-assistant sessions into a searchable ledger of spend, files touched, and what the assistant actually did. The exact adoption motive and repeat use are not yet verified.
in "analyze your history" tools, the top tier does not just tell you what you spent — it distills history into directly reusable assets (scripts, skills, instructions). That step from visibility to reusability is worth copying in any agent-workflow tool.
Decant itself is free and open source (Apache-2.0). Dosu's revenue sits in its main product; (agent knowledge infrastructure); the Decant blog post lands on "see what knowledge your agents; lack, then use Dosu to fill it." ① Whether stars clear 1,000 within three months and the HN attention converts to sustained; adoption — current public heat is low (10 points / 0 comments); ② Whether any user publicly reports that distilled scripts are actually being reused — it is; the only feature that differentiates it from a led…
Turns local coding-assistant sessions into a searchable ledger of spend, files touched, and what the assistant actually did.
It promises a simpler way to complete this job: Turns local coding-assistant sessions into a searchable ledger of spend, files touched, and what the assistant actually did. 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: “Turns local coding-assistant sessions into a searchable ledger of spend, files touched, and what the”. 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 local-first tool that turns the Claude Code and Codex session logs on your machine into a searchable, analyzable knowledge base: where tokens and cost go, how full the context windows run, which files and tools agents touch, full-text search across complete transcripts — plus distilling reusable workflow scripts from your command history. Zero outbound network calls at runtime; transcripts never leave the machine.
Maintained by Dosu (dosu.dev), self-described as "Knowledge Infrastructure for Agents." The
repo dosu-ai/decant is Apache-2.0, Bun + TypeScript; the most recent active contributor is
Taylor Dolezal (GitHub teedole). The pool lists devstein as builder, likely the repo's
creator; his relationship to Dosu could not be verified.
Read: Dosu builds agent infrastructure and runs agents daily, so this is dogfooding its own logs — Decant is both a community tool and an acquisition funnel for its main product. That "company core business + free open-source satellite" combo has steadier maintenance momentum than a purely personal project.
~/.claude/projects) and Codex (~/.codex)
logs into a single SQLite databaseWhat it deliberately does not do: no cloud, no network calls, no telemetry; because real transcripts can contain source code, prompts, credentials, and local paths, the project explicitly requires synthetic session data in issues and tests.
Reconstructing "what that agent session actually did and cost" used to mean digging through
scattered JSONL logs under ~/.claude/projects or trusting a vendor dashboard's single total.
Relationships between sessions, tool-call patterns, and "which files did agents keep re-reading
this week" could not be answered by eyeballing raw logs.
Decant replaces three chores: reading raw logs (→ one SQLite archive + full-text search), unclear cost attribution (→ token/cost/context broken down by dimension), and, most distinctively — turning past agent work into reusable assets (distill freezes a successful run into scripts/skills). That last one stops being accounting and starts being compounding.
Decant itself is free and open source (Apache-2.0). Dosu's revenue sits in its main product (agent knowledge infrastructure); the Decant blog post lands on "see what knowledge your agents lack, then use Dosu to fill it."
Read: the classic tool-as-funnel structure — the analyzer is free for installs and goodwill, and the money lives in the managed "make agents faster and cheaper" service. Judging Decant as a standalone business is meaningless; judging it as a funnel for Dosu is the real question.
| Dimension | Call |
|---|---|
| Founder-product fit | Dosu is itself a heavy agent user; the insight "agents keep relearning knowledge the company already has" comes straight from self-use |
| Product insight | Hits the same need as CodeBurn from a different angle — not just a ledger, but search and distillation, upgrading "analysis" into "reuse" |
| Execution quality | Local API with an OpenAPI contract, SLSA release verification, synthetic-data discipline — solid engineering norms |
| Timing | Appeared within days of CodeBurn; runaway AI cost is now a confirmed common pain and the category is forming |
Landing on the same need as CodeBurn is itself the evidence that the category exists.
Two products surfaced within the same week: CodeBurn (personal open source, cost dashboard across 40 tools with budget guards) and Decant (company open source, Claude Code/Codex session archive plus search and distill). The same pain picked independently by two parties is the strongest possible signal that runaway AI cost is a general problem.
Decant's most notable feature is distill. A ledger tells you where the money went; distill turns successful operation history directly into scripts, skills, and AGENTS.md sections — moving from "read the books" to "turn past work into future leverage." That is the line that separates it from CodeBurn, and the real landing point of the "knowledge infrastructure" thesis.
Two limits: coverage is only Claude Code and Codex, far narrower than CodeBurn's 40 tools; and as a funnel piece, its roadmap will follow Dosu's main-product strategy, not the standalone needs of Decant users.
① Whether stars clear 1,000 within three months and the HN attention converts to sustained adoption — current public heat is low (10 points / 0 comments) ② Whether any user publicly reports that distilled scripts are actually being reused — it is the only feature that differentiates it from a ledger ③ Whether tool coverage expands beyond Claude Code and Codex — coverage is the life-or-death line for this category
Product logic: in "analyze your history" tools, the top tier does not just tell you what you spent — it distills history into directly reusable assets (scripts, skills, instructions). That step from visibility to reusability is worth copying in any agent-workflow tool.
Unproven. The company backing is steadier than a solo project, distill is genuinely distinctive, but public heat is low, tool coverage is narrow, and feature evolution is held hostage by Dosu's main product. The category judgment — cost runaways are a common pain — was already proven by Decant and CodeBurn appearing in the same week; wait on the product itself.
05
Use these searches when the official site is missing or the current link is only a lead.