kikimimi

kikimimi records what your AI coding agents actually do — Claude Code today, Codex CLI too — through each agent’s own native mechanisms: hooks and OpenTelemetry for Claude Code, session-log tailing for Codex. No proxy, no TLS interception, no per-tool setup beyond one kikimimi init.

Everything lands in local Parquet first. Sharing anything past your own machine — a team dashboard, your own S3 bucket — is opt-in.

Why

Cost visibility

The invoice is the first signal for most people, and it shows up days late. kikimimi reads the same token/cost numbers Claude Code’s own OpenTelemetry export already carries and writes them locally as they happen — kikimimi query today is live, not a monthly surprise.

Thrash

A stuck agent gets creative: retrying the same failing tool call over and over, or working around a permission denial with bash. kikimimi query thrash finds both patterns per session. It’s a v0 proxy, not a certainty — see Quickstart for exactly what it does and doesn’t catch.

Context and schema tax

MCP servers you configured but never use still ship their schemas with every request. kikimimi query unused-mcp shows which ones are configured but idle; kikimimi query skills shows which skills actually get invoked (detecting unused skills the way unused-mcp does for MCP servers isn’t built yet). kikimimi query schema-tax estimates how much of a session’s first-turn context is fixed overhead rather than the actual prompt — labeled a proxy in its own output, because it is one.

Install

brew install isamisushi/tap/kikimimi

See Installation for the install script, platform support, updating, and the duckdb CLI that kikimimi query and kikimimi web need on PATH.

Quickstart

kikimimi init  # writes hooks + OTel env into your agent settings (backs up first);
               # also installs the daemon as a user service (launchd/systemd --user)
               # so it survives reboots and crashes -- kikimimi agent & still works too
kikimimi web   # local dashboard on 127.0.0.1 — nothing leaves your machine; needs duckdb on PATH
kikimimi query thrash      # stuck-agent incidents
kikimimi query unused-mcp  # context tax you pay for nothing

See Quickstart for what each command actually does. Existing Claude Code history shows up too — the first kikimimi agent start backfills sessions that already finished before this machine started collecting.

What it never does

Metadata only, by default and by schema: tool names, MCP server/tool, skill names, durations, success/failure, tokens, cost, session/repo identifiers. Prompt text, tool arguments, and command lines have columns reserved for them in the schema, but nothing populates those columns yet — opt-in content collection is a later stage. Anything sent to the hosted cloud has those columns nulled server-side too, as a second layer, not just a client promise. The hook shim always exits 0: kikimimi failing must never break or slow your agent.

All your machines, one place (free)

kikimimi login (GitHub device-code flow) syncs metadata-only events from every machine you use — laptops, VMs, CI — into your personal org on kikimimi.dev. Free for individuals. The local kikimimi web dashboard only ever shows this machine; the hosted one shows all of them. Nothing beyond metadata leaves the machine — see What it never does above.

Teams (optional)

Create a team org instead of using your personal one to add invite links, roles with audited admin drilldowns, and a per-machine repo allowlist. See Teams for the full picture.

Status

Early and moving fast. The kikimimi.v1 schema is additive-only but may still grow. Where kikimimi can’t measure something, it reports that as unknown rather than estimating it — the thrash and schema-tax queries call out their own known blind spots directly in their descriptions rather than hiding them.

Licensed Apache-2.0; only the hosted cloud service under crates/cloud/ is FSL-1.1-Apache-2.0 (converts to Apache-2.0 two years after each release).