Runtime reference

GitHub Copilot#

Copilot CLI keeps a full event log per session plus a SQLite usage ledger. Because Copilot bills in AI credits and the ledger records them exactly, cost here is vendor-billed truth rather than a per-token estimate. With the Pro extension 0.7.28 or later, Copilot Chat conversations in VS Code and VS Code Insiders are read too, as the same runtime with the surface vscode.

Where the data comes from#

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

• ~/.copilot/session-state/<uuid>/events.jsonl -- full event log
• ~/.copilot/session-state/<uuid>/workspace.yaml -- session metadata
• ~/.copilot/session-store.db -- SQLite index and usage ledger
• <VS Code user dir>/workspaceStorage/<id>/chatSessions/<sessionId>.jsonl -- VS Code Copilot Chat, a window with a folder open (Pro 0.7.28+)
• <VS Code user dir>/globalStorage/emptyWindowChatSessions/<sessionId>.jsonl -- VS Code Copilot Chat, a window with nothing open (Pro 0.7.28+)

Format. JSONL events ({type, data, id, timestamp, parentId}) plus a WAL-mode SQLite index.

Environment variables#

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

Tokens and cost#

Tokens. Real, from the assistant_usage_events ledger, with input, output, cache read, cache write and reasoning splits.

Cost. Vendor-billed: total_nano_aiu is the CLI's displayed credits times 1e9, and one credit is one US cent.

Sub-agents and delegation#

Custom agents run inside the parent session -- no child transcript is ever written. ClawMetry synthesises a child session per subagent.started, and the ledger's initiator='sub-agent' rows split child spend out of the parent rather than double-counting it.

Things worth knowing#

• input_tokens in the ledger already includes cache reads and output_tokens already includes reasoning tokens. ClawMetry reports uncached input plus a separate cache split so the UI does not double-count.
• VS Code Copilot Chat (Pro 0.7.28+). The VS Code user directory is ~/Library/Application Support/Code/User on macOS, $XDG_CONFIG_HOME/Code/User (default ~/.config) on Linux and %APPDATA%\Code\User on Windows; Insiders uses Code - Insiders. Only sessions answered by the GitHub Copilot Chat extension with at least one request are counted. Tokens come from the request's prompt and completion counts. Cost is the session's Copilot credits when VS Code records them, otherwise the sum of per-turn credits marked as a lower bound, otherwise unknown, never estimated; one credit is one US cent. VS Code stores no system prompt. JetBrains and browser Copilot Chat are not read.
• VS Code Copilot Chat sessions share the editor's process, so Guard refuses Pause, Stop and Kill on them (clawmetry 0.12.877 or later).
• GitHub organization Copilot usage (Pro 0.7.28+) is a separate, opt-in operator command, python -m clawmetry_pro.copilot_org_metrics, with a token you supply. It stores GitHub's organization and per-user metrics reports and AI-credit billing usage reports as consumption figures, never joined to local sessions and not an invoice.

Verify it is being read#

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

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

# 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.