---
title: Kimi
description: How ClawMetry observes Kimi — where its data lives, the environment variables that move it, what cost data is real, and the traps in its store.
eyebrow: Runtime reference
---

# Kimi

Kimi CLI and its successor Kimi Code CLI share a product line and a format but use different share directories. ClawMetry reads both and reports one runtime.

| | |
|---|---|
| **Runtime id** | `kimi` |
| **Category** | Terminal coding agent |
| **Vendor** | Moonshot AI |
| **Tier** | Paid tier (Starter and up) |
| **Upstream** | [github.com/MoonshotAI/kimi-cli](https://github.com/MoonshotAI/kimi-cli) |

## Where the data comes from

ClawMetry reads these locations, in order, and never writes to any of them.

- `$KIMI_SHARE_DIR or ~/.kimi` — Kimi CLI
- `~/.kimi-code` — Kimi Code CLI
- `<share>/sessions/<md5(work_dir)>/<session-uuid>/wire.jsonl` — the event log
- `<share>/sessions/…/context.jsonl, state.json, subagents/<id>/`
- `<share>/kimi.json` — work-dir registry; <share>/config.toml — models

**Format.** `wire.jsonl` is one record per line, optionally preceded by a protocol-version header. Envelopes include `TurnBegin`, `ContentPart`, `ToolCall`, `ToolResult` and `StatusUpdate`.

## Environment variables

| Variable | Effect |
|---|---|
| `KIMI_SHARE_DIR` | Kimi's own share-dir env |
| `CLAWMETRY_KIMI_ROOTS` | extra share dirs to consider |

See [Environment variables](/docs/config/environment/) for the variables that apply to every runtime.

## Tokens and cost

**Tokens.** Real: `StatusUpdate` carries the full cache split per step, plus context-window occupancy.

**Cost.** Derived — but note the model honesty caveat below.

:::note Why this section exists
Runtimes differ enormously in what they persist. Rather than showing a plausible number everywhere, ClawMetry reports what the store actually contains and marks the rest unknown. A cost of `$0.00` and a cost of *unknown* mean different things, and the dashboard keeps them different.
:::

## Sub-agents and delegation

Each sub-agent has its own directory with meta, context and wire logs.

## Things worth knowing

- Kimi persists a complete token breakdown but **not the model id** — no wire record carries it. The model is resolved from `config.toml`, which is the model configured *now*, so such sessions are marked `cost_status="estimated"` with `modelSource="config"`. No pricing match means cost is unknown, never a fabricated zero.
- The session directory's parent is `md5(work_dir)`, which is why project attribution comes from hashing each registered work dir in `kimi.json` rather than trying to invert a hash.

## Verify it is being read

```bash
# Is this runtime detected, and would this install unlock it?
clawmetry runtimes --json | jq '.runtimes[] | select(.id=="kimi")'

# Why is it locked, if it is?
clawmetry runtimes --why kimi

# Any sessions ingested yet?
curl -s 'http://localhost:8900/api/local/sessions?limit=5' \
  | jq '.rows[] | {session_id, agent_id, events, cost_usd}'
```

If detection succeeds but no sessions appear, work through [Troubleshooting](/docs/ops/troubleshooting/) — the usual causes are a relocated data directory (set the override above), a session limit that is hiding older runs, or a runtime that simply has not written anything yet.