§00Changelog
Every version, and what it changed for you.
63 releases since 25 Jul 2026. Breaking changes say what to do about them, and a version that rebuilds your index says so at the top.
Archis updates itself: chi update checks what is published, verifies the release hash before writing anything, and puts the previous binary back if the new one does not start. chi update --check installs nothing.
0.52.00.51.00.50.00.49.00.48.00.47.00.46.10.46.00.45.00.44.00.43.00.42.10.42.00.41.10.41.00.40.00.39.00.38.00.37.00.36.40.36.30.36.20.36.10.36.00.35.00.34.00.33.00.32.00.31.00.30.00.29.00.28.00.27.00.26.00.25.00.24.00.23.00.22.00.21.00.20.00.19.00.18.00.17.00.16.00.15.00.14.30.14.20.14.10.14.00.13.00.12.10.12.00.11.00.10.10.10.00.9.00.8.00.7.00.6.00.5.00.3.00.2.00.1.0
0.25.0
Forty commands are twenty-five, and everything you already type still works.
Before you upgrade — Nothing to change today. Your existing commands keep working and each one tells you its replacement the next time you run it —
chi agentsis the one to learn first.- NewWhat each agent costs you in tokens, on the repository and on the file. The dashboard could say which agent went through which file; nothing said what carrying it cost. Now it reads per agent: tokens shown, how much of that was exact source rather than a compact projection, what was held back behind handles, and the share of context that avoided. Opening a file answers the same question for that one file, including what a single look at it costs on average. Reads your agent made with its own tools stay a separate count on purpose — nobody counted their tokens, and averaging over both would invent a figure.
- ImprovedOne command per question, instead of one per subsystem. The surface had grown a command for every part of Archis rather than for every thing you want to know, and the duplicates were not subtle — two commands updated the binary, two listed your devices, and one described itself in its own help as a wrapper over another.
chi agentsis now the answer to "is Archis actually connected here": every agent, whether its CLI is present, whether the tools are registered, whether the hooks are installed, whether it has been briefed — one table, andchi agents onarranges all of it.chi refsreplaceschi callers,chi calleesandchi impact, and with no flag answers both cheap directions at once, because that is the question you actually have.chi runreplaceschi agent,chi updatereplaceschi self update,chi supportreplaceschi capabilitiesandchi coverage,chi symbol --sourcereplaceschi source,chi index --detectreplaceschi detect,chi dashboard --plainreplaceschi stats,chi doctor versionsreplaceschi versions, andchi init --machinereplaceschi setup. - ImprovedNothing you have already typed, scripted or put in CI stops working. Every older spelling still parses and still does exactly what it did — it is hidden from
--helpand prints one line to your terminal naming the command that replaced it. The hook entry points a harness invokes on your behalf are unchanged and stay silent, because they run once per tool call inside a live session. - ImprovedOne grammar for everything you can switch on and off:
on,off, and the bare command for where it stands —chi telemetry,chi license,chi keys,chi filter,chi agents. There were four spellings of this before, and five separatestatussubcommands the bare form now replaces.--json,--db,--sessions,--repoand--tokenizerare global and accepted on either side of the subcommand, so an invocation you already have keeps parsing whichever way you wrote it.chi agents advice offis new rather than renamed: turning the advice hook off had no command at all, and the only uninstall removed every hook — turning "stop commenting on my reads" into "stop recording anything". - ImprovedA dashboard ordered by the question you came with. The home screen leads with whether Archis is being used, by which agents and at what cost, and puts the repository grid under a machine-wide agents table; a repository page puts who worked here directly beneath whether Archis was used, ahead of the index figures. Every list is paged — including the short ones, because a table that starts paginating only once it is long changes shape exactly when there is most to read — and sessions, tool events, tools and languages now arrive in far larger slices.
- FixedA spent exact-source budget no longer reads to a model as "Archis is finished here". A session that hit the cap was told which call to use instead and nothing else, so the model concluded Archis had nothing left and went back to reading your files whole — the one outcome the cap exists to prevent. The message now says that only exact source is capped, which calls still answer and how much of what source was being read for they cover, and that a fresh session starts a fresh budget.
Run the current release
Archis Free is the whole local engine, with no card. Everything semantic runs on your machine — there is no endpoint that accepts source.