CClawMetryDocs

Runtime reference

Claude Code#

Claude Code writes a JSONL transcript per session under its projects tree. ClawMetry maps each turn to OpenTelemetry-shaped spans, so an assistant turn becomes an llm.call span, each tool use becomes a child span, and a Task becomes an agent.spawn — which is what makes the Agent Graph and the turn-anatomy views work.

Runtime idclaude_code
CategoryTerminal coding agent
VendorAnthropic
TierPaid tier (Starter and up)
Upstreamclaude.com/claude-code

Where the data comes from#

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

  • ~/.claude/projects/<project-slug>/<session-id>.jsonl — transcripts
  • $CLAUDE_CONFIG_DIR — when the CLI has been relocated
  • ~/Library/Application Support/Claude — Claude Desktop sidecar records (macOS; %APPDATA%\Claude on Windows, $XDG_CONFIG_HOME/Claude on Linux)

Format. One JSONL file per session; each line is a conversation turn with content blocks (text, thinking, tool_use, tool_result).

Environment variables#

VariableEffect
CLAUDE_CONFIG_DIRClaude Code's own config-dir override
CLAWMETRY_CLAUDE_DESKTOP_DIRDesktop state dir override
CLAWMETRY_FAMILY_SESSION_LIMIThow many recent sessions to ingest (default 50)
CLAWMETRY_DISABLE_CLAUDE_SESSION_INGESTturn the adapter off entirely

See Environment variables for the variables that apply to every runtime.

Tokens and cost#

Tokens. Full split per turn, including cache reads and writes.

Cost. Derived from the token split and model. Claude Code's native OpenTelemetry export also reports cost on both a log and a metric — ClawMetry counts it once.

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#

tool_use with name Task becomes an agent.spawn span and a child session, so sub-agent spend is attributed to the child rather than smeared across the parent.

Things worth knowing#

  • The CLI, the Desktop app and the VS Code extension all write into the same ~/.claude/projects tree. ClawMetry labels the surface each session came from so a Desktop conversation is not reported as a terminal run.
  • Sessions that OpenClaw spawned are ingested under the OpenClaw session id and skipped here — otherwise the same work appears twice.

Verify it is being read#

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

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

# 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 — 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.

Edit this page View as Markdown Report an issue
Cookie preferences