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

# Cursor

Cursor keeps chat in a VS Code-style SQLite key-value store, and its modern Agent mode writes a separate JSONL transcript under the home directory. ClawMetry reads both, because either one alone leaves a live session looking empty.

| | |
|---|---|
| **Runtime id** | `cursor` |
| **Category** | AI IDE (editor + CLI) |
| **Vendor** | Anysphere |
| **Tier** | Paid tier (Starter and up) |
| **Upstream** | [cursor.com](https://cursor.com) |

## Where the data comes from

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

- `~/Library/Application Support/Cursor/User/globalStorage/state.vscdb` — primary (Linux: ~/.config/Cursor/…)
- `…/workspaceStorage/<hash>/state.vscdb` — older per-workspace layout
- `~/.cursor/projects/<encoded-project>/agent-transcripts/<sid>/<sid>.jsonl` — Agent mode transcripts

**Format.** SQLite `cursorDiskKV` rows (`composerData:<id>`, `bubbleId:<composer>:<bubble>`) plus Anthropic-shaped JSONL for Agent sessions.

## Environment variables

| Variable | Effect |
|---|---|
| `CLAWMETRY_CURSOR_DB` | point at a specific state.vscdb |
| `CLAWMETRY_CURSOR_AGENT_ROOT` | Agent-transcript root override |
| `CLAWMETRY_CURSOR_CHATS_ROOT` | chat root override |

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

## Tokens and cost

**Tokens.** Cursor's `tokenCount` hint is not a billed total, and the backend is proprietary. On hosted VMs the loopback usage log is the only real token source.

**Cost.** Not on disk. Cursor is one of two runtimes on the token-blind allowlist for the hosted-VM usage log, which is why a Cursor row can show real cost on a managed node and none on a laptop.

:::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

Sub-agent transcripts live under a `subagents/` subtree of the session directory; the primary transcript is exactly `<sid>/<sid>.jsonl`.

## Things worth knowing

- Cursor is usually running, so recent writes sit in the `state.vscdb-wal` sidecar. Opening the database with `immutable=1` skips the WAL and returns a stale or empty view.
- A Cursor **CLI** session is a real process tree and can be paused or stopped. A Cursor **editor** conversation shares the one IDE process and cannot — the Guard tab says so per session rather than refusing the whole runtime.
- Agent-mode records carry no per-line timestamp; ClawMetry falls back to file mtime and offsets each message by a millisecond to keep order.

## Verify it is being read

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

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

# 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.