# letopis vs. your current stack — trackers, wikis, CI and kiro.dev > Every tool you already own answers one question and goes blind at the next. No single product covers the chain — just four categories of incumbent, each stopping exactly where the next question starts. Canonical page: https://dirahq.sh/letopis/compare ## The question | The question | You'd reach for | What it cannot see | letopis | | --- | --- | --- | --- | | How long did this actually take? | Toggl, Harvest, Clockify | Whether the timer was a human or a forgotten tab — and it never touches the artifact the hour produced. | dira | | Who — or what — wrote this code? | git blame, co-author trailers, Copilot metrics | The workstation. A trailer is a claim anyone can type; the code host cannot witness the session that produced the diff. | dira | | Why is it built this way? | Confluence, Notion, ADR folders, kiro.dev specs | Whether anyone followed it. A wiki page enforces nothing, and a generated spec is a plan the next agent run can quietly rewrite to agree with itself. | zavet | | Was this change properly governed? | CI, branch protection, compliance collectors | That a check ran is not that a human supervised. They record the gate, never the supervision behind the change passing through it. | proba | | What should the next one cost? | Spreadsheets, story points, gut feel | That its calibration died in 2024. Every input is a self-reported number from before agent leverage existed. | miara | ## Closest neighbour · kiro.dev **Kiro writes the spec before the work. letopis records what the work actually did.** Kiro is the one product on this page we would not call structurally blind. AWS built it on the same premise we did — that intent belongs in the repo, versioned, beside the code — and it is genuinely good at the half it owns. It is upstream of letopis, not instead of it. A spec is a stated *intention*; a chronicle is an *observation*. Only one of those survives contact with an agent. **Kiro · spec-first, in the IDE** - `.kiro/specs` — requirements, design and tasks, authored before the code and tracked against it. - Steering files apply team conventions to every agent action, first-class and versioned. - Hooks fire on save and commit — inside Kiro. - ✕ **Sees only work done in Kiro.** The editor is the boundary. Your teammate on Claude Code is invisible to it. - ✕ **Its artifacts are generated, unsigned and rewritable.** Refine the design and the next run syncs over it. Nothing stops an agent from making the spec agree with the code it just wrote. - ✕ **Cannot witness the merge.** No CI state at merge, no non-author review, no supervision minutes — the IDE is on the wrong side of the pull request. - ✕ **No hours, no client, no invoice.** It prices agent credits, not the work. Nothing in it tells you what the feature cost or what the next one should. **letopis · record-first, around the repo** - Decisions captured as a byproduct of the work, then **guarded** — an agent that contradicts one has to surface it, not overwrite it. - Every harness the developer actually used — Claude Code, Codex, Cursor, OpenCode. **The git remote is the boundary, not the editor.** - Witnesses the merge: CI at the merge commit, review by a non-author, supervision minutes, no branch protection bypassed. - Time, compute cost, client, invoice and the next quote all hang off the same signed session. - ✓ Signed at capture on the workstation, immutable after — which is what makes it evidence rather than documentation. - And what letopis *doesn't* do: it will not write your spec, plan your tasks or steer your agent. Kiro is better at that than we intend to be. **Run both.** A `.kiro/spec` is perfectly good evidence of intent, and zavet will cite it like any other decision record — a change that implements one reaches 999 fineness the same way. What Kiro cannot do is tell you whether anybody followed it. ## And then the real problem **Buy all four categories and the interesting questions are still unanswerable.** Because they are questions *across* links, and five vendors share no key. letopis has exactly one: the signed session every other record hangs off. This is the only thing letopis sells that cannot be assembled from parts. Anyone can build a timer, a wiki or a CI check — nobody else can join them, because nobody else signs the session all four hang off. ### What did decision D-0042 cost us? **Your stack today** | Source | Holds | | --- | --- | | Confluence | the decision page — no field a timer knows about | | Toggl | 3h 12m against "API work" — no decision reference | | Provider console | token spend by day, by nobody in particular | ✕ 3 systems · 0 shared keys · unanswerable **letopis** | Source | Holds | | --- | --- | | zavet · D-0042 | the decision, its guards, the paths it owns | | dira · s-2481, s-2490 | the sessions that referenced it, signed | | the join | decision ↔ session, same key | ✓ 3h 12m engaged · 14h 02m agent · 612k tok — and the guard has blocked 4 reverts of it since ### Did a human supervise PR #4172, and for how long? **Your stack today** | Source | Holds | | --- | --- | | GitHub | a review approval and a merge timestamp | | Co-author trailer | a line of text anyone can type | | Nothing | measures how long a human actually watched | ✕ the workstation is invisible to the code host **letopis** | Source | Holds | | --- | --- | | proba · hallmark #4172 | CI green, non-author review, fineness 750 | | dira · signed session | harness identity and engaged minutes | | the join | merge ↔ session, cryptographic | ✓ 47m engaged against 4h 20m of agent time, reviewed by a non-author — recorded, not claimed ### Is this quote comparable to the last four? **Your stack today** | Source | Holds | | --- | --- | | Your invoice history | hours you typed in yourself | | A spreadsheet | a velocity calibrated before agents existed | | Gut feel | wrong by ~10× depending on work type | ✕ your own numbers are the ones you cannot verify **letopis** | Source | Holds | | --- | --- | | miara · 4 confirmed comparables | past deliverables you picked as similar | | dira · anchored deliveries | the real engaged hours behind each one | | proba · fineness filter | only work governed at 750 or better counts | ✓ n = 4 at 750+ · 2.1–4.8h engaged · spread 2.3× — or it refuses to answer at all ## Five tools, or one chain. dira is free forever for solo work and open source. Everything else is written on top of it. - [Start tracking free](https://dirahq.sh/letopis#pricing) - [The AI Act case](https://dirahq.sh/letopis/ai-act) --- See also [/letopis.md](https://dirahq.sh/letopis.md), [/letopis-ai-act.md](https://dirahq.sh/letopis-ai-act.md), [/compare.md](https://dirahq.sh/compare.md).