0.7.1 - 2026-09-08

Release Notes

Added

• Codex subscription usage is collected automatically: the kikimimi agent daemon
  refreshes the Codex snapshot every 5 minutes from the signed-in Codex profile
  (CODEX_HOME or ~/.codex), so the Overview and kikimimi usage stay current without a
  manual run. --account and --profile on kikimimi usage codex are now optional — the
  label defaults to the authenticated ChatGPT account ID, read from auth.json identity
  fields only (credentials are never copied into snapshots or logs). A snapshot is refused
  when the account changes mid-collection or the app-server's accountId disagrees with
  the local profile, so usage can never be attributed to the wrong subscription. Profiles
  with no ChatGPT sign-in (API-key or keychain-backed) are skipped unless --account is
  given, failed refreshes keep the previous observation and retry, and
  KIKIMIMI_NO_CODEX_USAGE=1 turns the collector off. Local only, as before.

Fixed

• kikimimi self-update now restarts the installed daemon service. The agent handles
  SIGTERM as a clean exit, so KeepAlive / Restart=on-failure never brought it back and
  the machine stopped collecting until the next manual start; the update now explicitly
  restarts through launchctl kickstart / systemctl --user restart, keeping the service's
  environment and avoiding an unmanaged duplicate. A failed restart, or a failed manual
  daemon respawn, is reported instead of printing success.

Install kikimimi 0.7.1

Install prebuilt binaries via shell script

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/isamisushi/kikimimi/releases/download/v0.7.1/kikimimi-installer.sh | sh

Install prebuilt binaries via Homebrew

brew install isamisushi/tap/kikimimi

Download kikimimi 0.7.1

File	Platform	Checksum
kikimimi-aarch64-apple-darwin.tar.xz	Apple Silicon macOS	checksum
kikimimi-x86_64-apple-darwin.tar.xz	Intel macOS	checksum
kikimimi-aarch64-unknown-linux-gnu.tar.xz	ARM64 Linux	checksum
kikimimi-x86_64-unknown-linux-gnu.tar.xz	x64 Linux	checksum
kikimimi-aarch64-unknown-linux-musl.tar.xz	ARM64 MUSL Linux	checksum
kikimimi-x86_64-unknown-linux-musl.tar.xz	x64 MUSL Linux	checksum

0.7.0 - 2026-09-08

Release Notes

Added

• Subscription usage by account (local Overview and kikimimi usage): separate
  Claude/Codex account labels, current usage windows, reset times, and freshness.
  Claude status-line input and Codex App Server supply the snapshots; unknown
  windows stay unknown, and concurrent updates keep the latest value without summing.
  See setup. Local only; labels are
  explicitly assigned and subscription data is not synced to the cloud.
• Models page (web, /models; GET /web/q/models?days=N on both kikimimi cloud and the local
  kikimimi web daemon, KKM-34): which model, at which effort, burned how many tokens — one row per
  (model, effort) with API requests, API errors, sessions, the share of requests and tokens that
  ran inside subagents, input / output / cache read / cache write / reasoning tokens and cost, plus
  a daily stacked chart by model. Org-wide aggregate only (every role), api.request based; a
  request seen by both OTel and the transcript backfill is counted once (OTel wins per session,
  the subagents rule). effort is Claude Code's own per-request field and is NULL for its
  internal helper calls; model is unknown when a source had none.
• Session detail — model × effort: a per-session Models table (same columns as above, minus
  sessions), models / efforts / model_source on every subagent row (what the subagent actually
  ran on — its own transcript rows, else attributed from the session's OTel api.request rows by
  agent type and time window, since live hooks carry no model and OTel no agent_id),
  effort on every event row, and efforts in the summary. All appended columns, so existing
  consumers of /web/q/session are unaffected.
• Local web daemon: GET /web/me now returns the full SessionInfo shape (orgs,
  active_org, …); the old {email, org_id} body crashed the SPA's layout on every page.
• Codex reasoning_effort (turn_context, top-level or collaboration_mode.settings) now
  fills effort on Codex turn / api.request rows; null stays NULL.

Install kikimimi 0.7.0

Install prebuilt binaries via shell script

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/isamisushi/kikimimi/releases/download/v0.7.0/kikimimi-installer.sh | sh

Install prebuilt binaries via Homebrew

brew install isamisushi/tap/kikimimi

Download kikimimi 0.7.0

File	Platform	Checksum
kikimimi-aarch64-apple-darwin.tar.xz	Apple Silicon macOS	checksum
kikimimi-x86_64-apple-darwin.tar.xz	Intel macOS	checksum
kikimimi-aarch64-unknown-linux-gnu.tar.xz	ARM64 Linux	checksum
kikimimi-x86_64-unknown-linux-gnu.tar.xz	x64 Linux	checksum
kikimimi-aarch64-unknown-linux-musl.tar.xz	ARM64 MUSL Linux	checksum
kikimimi-x86_64-unknown-linux-musl.tar.xz	x64 MUSL Linux	checksum

0.6.1 - 2026-09-08

Release Notes

Added

• Session detail page (web, /sessions/<id>; GET /web/q/session?session_id=…&events_limit=N
  on both kikimimi cloud and the local kikimimi web daemon): click a session id on the Sessions
  or Subagents page to drill into one session — header numbers (duration, events, turns, tool
  calls, failures, denied, API requests, compactions, subagents, tokens incl. cache, cost, models,
  configured MCP servers / skills), an activity timeline (events per bucket, main vs. subagent,
  failures marked; bucket width chosen from the session's span and reported as bucket_ms), a
  per-tool table with p50/p95 and total duration and how many calls came from subagents, a
  per-subagent table (type, start, duration, tool calls, failures, estimated tokens — unknown
  when Claude Code reported none, never 0), and a filterable chronological event list (metadata
  only, first events_limit rows). Hook/OTel tool.result pairs are deduped once everywhere.
  Same role gate as Sessions: a team member only reaches their own sessions (404 otherwise), an
  admin/owner's request leaves a session_drilldown audit row naming the session.

• Onboarding funnel (KKM-21, architecture.md §15-15 / §12 Stage 0 "install → first insight
  in 2 minutes"): the cloud records the first time each host reached login_started,
  login_done and first_events, and each account reached first_insight (web Overview or
  kikimimi query --cloud), in a new funnel_steps table (migration 0014_funnel_steps;
  backfilled from existing devices and events). Opaque ids and timestamps only — no email,
  hostname or content; see Privacy.
  GET /web/q/funnel?days=N returns the per-step host counts for that cohort, the median
  minutes from login_started to first_events, and 30-day retention (hosts old enough to
  have a 30th day vs. hosts still authenticating after it, from devices). Two scopes that
  follow the role model: scope=org (default) is the active org's machines, for its admin/owner,
  shown on the Team page; scope=all is every org on the deployment and needs the new
  accounts.operator flag (set in SQL by whoever runs the instance — no HTTP path grants it;
  GET /web/me reports it). KIKIMIMI_FUNNEL=0 disables recording.

Fixed

• Empty agent_type / agent_id from Claude Code hooks are stored as NULL. Claude Code
  2.1.x sends agent_type: "" on some SubagentStop payloads, which surfaced as a blank entry in
  the Subagents page's agent-types column (,general-purpose). The hook adapter now treats ""
  as absent for both fields.

Install kikimimi 0.6.1

Install prebuilt binaries via shell script

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/isamisushi/kikimimi/releases/download/v0.6.1/kikimimi-installer.sh | sh

Install prebuilt binaries via Homebrew

brew install isamisushi/tap/kikimimi

Download kikimimi 0.6.1

File	Platform	Checksum
kikimimi-aarch64-apple-darwin.tar.xz	Apple Silicon macOS	checksum
kikimimi-x86_64-apple-darwin.tar.xz	Intel macOS	checksum
kikimimi-aarch64-unknown-linux-gnu.tar.xz	ARM64 Linux	checksum
kikimimi-x86_64-unknown-linux-gnu.tar.xz	x64 Linux	checksum
kikimimi-aarch64-unknown-linux-musl.tar.xz	ARM64 MUSL Linux	checksum
kikimimi-x86_64-unknown-linux-musl.tar.xz	x64 MUSL Linux	checksum

0.6.0 - 2026-09-07

Release Notes

Added

• Backfilled Claude Code sessions now carry configured_mcp_servers (from the MCP tool names
  the transcript lists — claude.ai connectors included) and a new additive configured_skills
  column (Claude Code's own skill listing; migration 0013_configured_skills) on their
  session.end, so unused-mcp, mcp-tax and the unused_mcp_server pattern no longer fall
  back to the 30-day proxy for them (KKM-18). Where a session has both the hook's session.start
  snapshot and the transcript's, the transcript's wins. New named query unused-skills (local
  and cloud) and a "Configured vs invoked" table on the Skills page (/web/q/unused-skills):
  skills Claude Code listed per session versus skills invoked; hooks-only sessions contribute
  nothing to "configured" rather than a directory guess.

• Codex session.start rows now carry configured_mcp_servers, read from the
  [mcp_servers.<name>] tables of $CODEX_HOME/config.toml (names only), so unused-mcp
  and the unused_mcp_server pattern cover Codex sessions (KKM-16). Codex tool.denied is
  still not emitted: the installed codex-cli's approval vocabulary was confirmed but no
  denial record was observed on disk, and kikimimi does not guess record shapes — see
  How it works.

• Coverage (KKM-17, architecture.md §7.1 "数字の信頼度を隠さない"): /web/q/coverage
  (local and cloud) returns the counts behind every missing-data rate — sessions without
  token usage, hook↔OTel tool_use_id match, what the dedup folded away, subagents without
  usage, events without an account, hosts silent for a day — and the web UI shows them as a
  badge in the top bar on every page plus a "Data coverage" panel on Overview. The hook shim
  now records its own latency (~/.kikimimi/shim-latency.log, rotated) and kikimimi status
  prints the p50/p99. Measured numbers are published in
  How it works.

• Subagent attribution (KKM-15, architecture.md §7.2 subagent_fanout). Three additive
  kikimimi.v1 columns — agent_id, agent_type, query_source (cloud migration
  0012_subagent_attribution) — carry which Agent-tool subagent a row belongs to, from the
  hook payload's agent_id/agent_type, the transcript's agentId/isSidechain, and OTel's
  query_source/agent.name. session_id stays the parent's, so every existing per-session
  view already includes the subagents. kikimimi init now also registers the SubagentStart
  hook (subagent.start event); the transcript backfill reads the sidechain transcripts under
  <session>/subagents/ (Workflow-launched ones included) and emits subagent.stop at their
  end instead of session boundaries. New named query subagents (local and cloud) and a
  Subagents web page (/web/q/subagents): per session, subagent count, types, tool calls,
  time share, token share where usage was recorded, and subagents_with_usage — the honest
  count of subagents that could be priced at all (Claude Code still loses subagent usage
  upstream: anthropics/claude-code #83430, #88107).

• context_bloat now compares requests within one conversation stream (main, or one subagent)
  instead of across the whole session. Parallel subagents alternating 40k/90k/45k/91k were the
  main false positive: on one machine's real data the same-model rule over OTel rows flagged 30
  jumps in one session; the per-stream rule over a full transcript backfill of that machine
  leaves 16 in the busiest session and 27 across 60 sessions.

• A zero-row schema stub (~/.kikimimi/data/events/dt=_schema/kikimimi.v1-<n>.parquet),
  written by the daemon at startup and by kikimimi query, so local DuckDB queries can name
  a column that this machine's older Parquet predates (DuckDB's union_by_name takes the
  union of every file's columns). Without it, query patterns on pre-upgrade data failed with
  "Referenced column not found" until the first post-upgrade flush.

• Persisted struggle-pattern detection (architecture.md §7.2, KKM-9). The cloud runs a
  background scanner (PATTERN_SCAN_INTERVAL_SECS, default 300) that writes one row per
  incident into a new org-scoped pattern_hits table (migration 0010_pattern_hits):
  mcp_bypass, deny_detour, retry_spiral, permission_denied_loop, context_bloat,
  long_tool_tail and unused_mcp_server, each attributed to a subject (the MCP
  server or tool) and priced with wasted_tokens_est (input+output tokens of the session's
  OTel api.request rows inside the incident window; NULL when unknown). Days are rescanned
  while events arrive until 72 hours after they end (PATTERN_WATERMARK_HOURS), then frozen;
  late arrivals are counted, not folded in. New named query patterns on both sides:
  kikimimi query patterns recomputes the same rows live from local Parquet,
  --cloud / GET /v1/query/patterns reads the table.

• Struggles page (web, local and hosted) and /web/q/patterns / /web/q/pattern-hits
  (KKM-11, architecture.md §7.2 "組織横断ランキング"): the detected patterns aggregated per
  (pattern, subject) — the MCP server or tool to fix, never a person — ranked by
  wasted_tokens_est × sessions; unpriced rows show "unknown" and sort last rather than
  pretending to be zero. Clicking a row lists the sessions behind it (metadata only). In a
  team org a member below admin only sees their own sessions in the drilldown, and an
  admin/owner drilldown writes an audit_log row, exactly like the Sessions page.

• Before/after on the Struggles page (KKM-12, architecture.md §7.3): each ranking row
  opens a per-day trend (sessions, sessions hit, hit rate — the §7.3 KPI — incidents, wasted
  tokens; /web/q/pattern-timeline) and an "improvement mark" (/web/marks, migration
  0011_improvement_marks; locally ~/.kikimimi/marks.json): "we changed this subject on
  ". With a mark the page shows the hit rate and cost before vs. after it. In a team org,
  recording or removing a mark needs admin or owner.

• Docs: The improvement loop — the
  runbook for running detect → fix → before/after with one team, what each pattern means for
  its subject, and how to sample false positives/negatives from the drilldown.

• mcp-tax named query (local and cloud; architecture.md §7.2 schema_tax, KKM-13): the
  per-session fixed context (schema-tax's first_input_tokens) allocated to the MCP servers in
  each session's configured_mcp_servers snapshot — fixed_tokens_est (equal split, paid per
  request), unused_tokens_est (the share carried by sessions that never called the server) and
  marginal_first_tokens_est (median with minus median without the server, when the org's
  configs give a contrast). The scanner persists the same allocation as unused_mcp_server
  pattern hits, one per (session, never-called server).

Install kikimimi 0.6.0

Install prebuilt binaries via shell script

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/isamisushi/kikimimi/releases/download/v0.6.0/kikimimi-installer.sh | sh

Install prebuilt binaries via Homebrew

brew install isamisushi/tap/kikimimi

Download kikimimi 0.6.0

File	Platform	Checksum
kikimimi-aarch64-apple-darwin.tar.xz	Apple Silicon macOS	checksum
kikimimi-x86_64-apple-darwin.tar.xz	Intel macOS	checksum
kikimimi-aarch64-unknown-linux-gnu.tar.xz	ARM64 Linux	checksum
kikimimi-x86_64-unknown-linux-gnu.tar.xz	x64 Linux	checksum
kikimimi-aarch64-unknown-linux-musl.tar.xz	ARM64 MUSL Linux	checksum
kikimimi-x86_64-unknown-linux-musl.tar.xz	x64 MUSL Linux	checksum

0.5.1 - 2026-09-07

Release Notes

Patch release. Re-running kikimimi init on 0.5.0 could move the OTLP port out from under
Claude Code; kikimimi flush now waits for uploads to finish; licensing is split so
everything you install is Apache-2.0.

Changed

• License split: the repository root is now Apache-2.0 (agent daemon, CLI, adapters,
  kikimimi.v1 schema, detection SQL, local web UI, docs). Only crates/cloud/ (the
  hosted cloud service) stays FSL-1.1-Apache-2.0, in crates/cloud/LICENSE.md. Previous
  releases were FSL-1.1-Apache-2.0 as a whole while Cargo.toml claimed Apache-2.0; the
  two now agree. See NOTICE.

Fixed

• kikimimi flush now waits until the daemon has flushed every sink and reports the
  outcome (flush ok file=1 s3=1, or flush error s3: <reason> with exit code 1).
  Before, it printed flush acked by daemon: true the moment the control byte was
  written, so an s3 upload still sitting in its retry loop (up to 3 x 60 s) was
  indistinguishable from a flush that never ran, and a container's postStopCommand
  could exit before the upload finished (GitHub #1). Against a daemon older than this
  release it still returns immediately and says so.
• kikimimi init no longer mistakes its own running daemon for a foreign process on the
  OTLP port. Re-running init while a 0.4.x kikimimi agent & (or the installed service)
  was still bound to 4318 picked a random alternate port, persisted it in config.json,
  and left OTEL_EXPORTER_OTLP_ENDPOINT at the old value ("differs, leaving unchanged"),
  so Claude Code kept exporting to a port nobody listened on. init now keeps the port
  when the daemon behind the control socket reports it in state.json, and when the port
  does legitimately change it rewrites the endpoint a previous init wrote instead of
  warning about it (a user-customized endpoint is still left alone).
  If you hit this on 0.5.0: set otlp_port back to 4318 in ~/.kikimimi/config.json and
  restart the service (systemctl --user restart kikimimi-agent / launchctl kickstart -k gui/$(id -u)/dev.kikimimi.agent), or run kikimimi init again with this fix.

Install kikimimi 0.5.1

Install prebuilt binaries via shell script

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/isamisushi/kikimimi/releases/download/v0.5.1/kikimimi-installer.sh | sh

Install prebuilt binaries via Homebrew

brew install isamisushi/tap/kikimimi

Download kikimimi 0.5.1

File	Platform	Checksum
kikimimi-aarch64-apple-darwin.tar.xz	Apple Silicon macOS	checksum
kikimimi-x86_64-apple-darwin.tar.xz	Intel macOS	checksum
kikimimi-aarch64-unknown-linux-gnu.tar.xz	ARM64 Linux	checksum
kikimimi-x86_64-unknown-linux-gnu.tar.xz	x64 Linux	checksum
kikimimi-aarch64-unknown-linux-musl.tar.xz	ARM64 MUSL Linux	checksum
kikimimi-x86_64-unknown-linux-musl.tar.xz	x64 MUSL Linux	checksum

0.5.0 - 2026-09-04

Release Notes

Upgrade note: re-run kikimimi init after updating (brew upgrade kikimimi, or
kikimimi self-update for shell-installer installs). init installs the daemon as a user
service and mints the OTLP bearer token. Until you do, the daemon keeps running the old way,
the local OTLP receiver stays unauthenticated (fail-open) and kikimimi status warns about it.

• If you started kikimimi agent by hand (nohup, a shell alias, a login item), init stops
  that process before installing the service, so the service takes over cleanly. Remove the
  alias or login item afterwards, or the next login starts a second daemon that exits on the
  socket liveness check.
• Claude Code sessions that were already open when you ran init still export OTel without
  the new header, so their telemetry is rejected until you restart them. kikimimi status
  warns with the rejected count.

Added

• kikimimi service install|uninstall|status: runs the daemon as a user service
  (macOS LaunchAgent / Linux systemd --user) so it survives reboots and crashes.
  kikimimi init installs and starts it (--no-service to skip, --dry-run to preview),
  kikimimi uninstall removes it, kikimimi status reports it, and self-update hands
  the restart to the service manager.
• Per-install bearer token on the local OTLP receiver: init mints it once (re-running
  init keeps it), writes OTEL_EXPORTER_OTLP_HEADERS into the Claude Code settings and
  persists it in config.json. The daemon requires Authorization: Bearer on /v1/logs,
  /v1/metrics and /v1/traces, reloads the token on the r control byte and counts
  rejections; kikimimi status warns when the receiver is unauthenticated or has rejected
  requests.
• configured_mcp_servers on Claude Code session.start (additive kikimimi.v1 column):
  the names of the MCP servers configured in ~/.claude/settings.json, ~/.claude.json
  (top level and the per-project entry) and <cwd>/.mcp.json. The cloud unused-mcp
  query and /web/q/unused-mcp use that snapshot to list servers that are configured but
  never called, even with zero events; the web endpoint carries a configured_from_snapshot
  flag and, on the cloud, falls back to the 30-day observed set for sessions without a
  snapshot. The web MCP page shows an Unused badge and a Configured column. Cloud
  migration 0009_configured_mcp_servers. (kikimimi query unused-mcp on a local install
  keeps reading the live config files.)
• One-shot backfill of Claude Code transcripts that ended before collection began
  (~/.claude/projects, metadata only, source=log). The boundary is the earliest local
  dt= partition, else the first daemon start, and is persisted once. Progress is
  checkpointed per batch in ~/.kikimimi/claude-backfill.json; the handoff to the cloud
  and s3 sinks is throttled so a long history drains gradually instead of overflowing the
  pending buffer. Opt out with claude_backfill = false in config.json or
  KIKIMIMI_NO_CLAUDE_BACKFILL=1. kikimimi status shows a claude backfill block.

Changed

• MCP servers whose name contains playwright, browser, chrome, webfetch or
  puppeteer (case-insensitive; covers Playwright, claude-in-chrome and puppeteer) are
  classified as tool_kind = browser, so bypass, thrash (deny_detour) and reach see
  browser detours that travel over MCP. mcp_server / mcp_tool are still set, so MCP
  health and unused-mcp are unaffected.
• kikimimi init prints a NOTE when the duckdb CLI is missing; the README and docs
  disclose the dependency. kikimimi query skills is documented as reporting invoked
  skills only (unused-skill detection is not built).

Fixed

• Hook and OTel rows for the same tool.result are deduplicated in every aggregation
  (tools, mcp, skills, today, bypass, thrash, sessions), in local DuckDB and
  in the cloud SQL. The OTel row wins, the hook row is the fallback.
• Cloud: the kikimimi_app role password is reconciled with KIKIMIMI_APP_DB_PASSWORD
  on every boot (#3).
• The "another instance is already listening" startup error no longer prints the
  kikimimi agent: prefix twice.

Install kikimimi 0.5.0

Install prebuilt binaries via shell script

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/isamisushi/kikimimi/releases/download/v0.5.0/kikimimi-installer.sh | sh

Install prebuilt binaries via Homebrew

brew install isamisushi/tap/kikimimi

Download kikimimi 0.5.0

File	Platform	Checksum
kikimimi-aarch64-apple-darwin.tar.xz	Apple Silicon macOS	checksum
kikimimi-x86_64-apple-darwin.tar.xz	Intel macOS	checksum
kikimimi-aarch64-unknown-linux-gnu.tar.xz	ARM64 Linux	checksum
kikimimi-x86_64-unknown-linux-gnu.tar.xz	x64 Linux	checksum
kikimimi-aarch64-unknown-linux-musl.tar.xz	ARM64 MUSL Linux	checksum
kikimimi-x86_64-unknown-linux-musl.tar.xz	x64 MUSL Linux	checksum

v0.4.0

Install kikimimi 0.4.0

Install prebuilt binaries via shell script

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/isamisushi/kikimimi/releases/download/v0.4.0/kikimimi-installer.sh | sh

Install prebuilt binaries via Homebrew

brew install isamisushi/tap/kikimimi

Download kikimimi 0.4.0

File	Platform	Checksum
kikimimi-aarch64-apple-darwin.tar.xz	Apple Silicon macOS	checksum
kikimimi-x86_64-apple-darwin.tar.xz	Intel macOS	checksum
kikimimi-aarch64-unknown-linux-gnu.tar.xz	ARM64 Linux	checksum
kikimimi-x86_64-unknown-linux-gnu.tar.xz	x64 Linux	checksum
kikimimi-aarch64-unknown-linux-musl.tar.xz	ARM64 MUSL Linux	checksum
kikimimi-x86_64-unknown-linux-musl.tar.xz	x64 MUSL Linux	checksum

v0.3.0

Install kikimimi 0.3.0

Install prebuilt binaries via shell script

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/isamisushi/kikimimi/releases/download/v0.3.0/kikimimi-installer.sh | sh

Install prebuilt binaries via Homebrew

brew install isamisushi/tap/kikimimi

Download kikimimi 0.3.0

File	Platform	Checksum
kikimimi-aarch64-apple-darwin.tar.xz	Apple Silicon macOS	checksum
kikimimi-x86_64-apple-darwin.tar.xz	Intel macOS	checksum
kikimimi-aarch64-unknown-linux-gnu.tar.xz	ARM64 Linux	checksum
kikimimi-x86_64-unknown-linux-gnu.tar.xz	x64 Linux	checksum
kikimimi-aarch64-unknown-linux-musl.tar.xz	ARM64 MUSL Linux	checksum
kikimimi-x86_64-unknown-linux-musl.tar.xz	x64 MUSL Linux	checksum

v0.2.0

Install kikimimi 0.2.0

Install prebuilt binaries via shell script

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/isamisushi/kikimimi/releases/download/v0.2.0/kikimimi-installer.sh | sh

Install prebuilt binaries via Homebrew

brew install isamisushi/tap/kikimimi

Download kikimimi 0.2.0

File	Platform	Checksum
kikimimi-aarch64-apple-darwin.tar.xz	Apple Silicon macOS	checksum
kikimimi-x86_64-apple-darwin.tar.xz	Intel macOS	checksum
kikimimi-aarch64-unknown-linux-gnu.tar.xz	ARM64 Linux	checksum
kikimimi-x86_64-unknown-linux-gnu.tar.xz	x64 Linux	checksum
kikimimi-aarch64-unknown-linux-musl.tar.xz	ARM64 MUSL Linux	checksum
kikimimi-x86_64-unknown-linux-musl.tar.xz	x64 MUSL Linux	checksum

v0.1.1

Install kikimimi 0.1.1

Install prebuilt binaries via shell script

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/isamisushi/kikimimi/releases/download/v0.1.1/kikimimi-installer.sh | sh

Install prebuilt binaries via Homebrew

brew install isamisushi/tap/kikimimi

Download kikimimi 0.1.1

File	Platform	Checksum
kikimimi-aarch64-apple-darwin.tar.xz	Apple Silicon macOS	checksum
kikimimi-x86_64-apple-darwin.tar.xz	Intel macOS	checksum
kikimimi-aarch64-unknown-linux-gnu.tar.xz	ARM64 Linux	checksum
kikimimi-x86_64-unknown-linux-gnu.tar.xz	x64 Linux	checksum
kikimimi-aarch64-unknown-linux-musl.tar.xz	ARM64 MUSL Linux	checksum
kikimimi-x86_64-unknown-linux-musl.tar.xz	x64 MUSL Linux	checksum