Runtime reference
Grok Bot#
Grok Bot is not the Grok Build CLI. It is a set of persistent named bots, each on its own cloud VM with a browser, filesystem and terminal, driven from a desktop client. The client is an Electron app built by Anysphere (Cursor's company), which is why its storage idioms rhyme with Cursor's and why a Cursor agent can appear inside a Grok Bot transcript.
| Runtime id | grok_bot |
| Category | Cloud agent, desktop client |
| Vendor | xAI |
| Tier | Paid tier (Starter and up) |
| Upstream | x.ai/bot |
Where the data comes from#
ClawMetry reads these locations, in order, and never writes to any of them.
~/Library/Application Support/Grok Bot/sand-client-persistence/<base32>.blob— conversations (macOS)%APPDATA%\Grok Bot\sand-client-persistence/— same, on Windows~/.config/Grok Bot/sand-client-persistence/— same, elsewhere~/.grokbot/— the local-exec control plane, NOT under the data root
Format. Each .blob filename is the lowercase base32 of its slice key with padding stripped; the body is plaintext JSON {"schemaVersion", "value"}. Nothing is encrypted, so no key material is needed and none is touched.
Environment variables#
| Variable | Effect |
|---|---|
CLAWMETRY_GROK_BOT_DATA_ROOT | conversation data root override |
CLAWMETRY_GROK_BOT_HOME | control-plane root override (~/.grokbot) |
See Environment variables for the variables that apply to every runtime.
Tokens and cost#
Tokens. None. An exhaustive walk of the store for token, usage, model, price and credit keys returns nothing — not a zero, not a null; the fields do not exist. Inference happens on the cloud VM and the metering never comes back to the client.
Cost. Unavailable. cost_usd is None with cost_status="unavailable" and the runtime does not claim the COST capability. A dollar figure here would be invented from nothing.
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#
A cursor-agent transcript entry records a handoff to a Cursor background agent by id and title. That is a link to another runtime's work, not a sub-agent tree of its own.
Things worth knowing#
- A
~/.grok-shaped check finds neither root. Grok Build (grok) and Grok Bot are different products that share a brand. - A
send-messageentry reads like an outbound queue and is not — it is the bot's own reply. Treating it as an outbox drops every assistant turn and leaves a transcript of the user talking to themselves. local-tool-permissionis the one genuine governance signal: the bot asks to act on YOUR machine, recordingaction,targetandstatus.status="always"is a standing grant that outlives the one ask.- The ask is the only tool signal persisted locally. There is no execution record, so ClawMetry emits the ask as a tool call and never fabricates a result.
- One desktop process serves every bot, so there is no per-session control and there never can be. Pause, stop and kill are refused with a reason rather than offered as buttons that do nothing.
Verify it is being read#
# Is this runtime detected, and would this install unlock it?
clawmetry runtimes --json | jq '.runtimes[] | select(.id=="grok_bot")'
# Why is it locked, if it is?
clawmetry runtimes --why grok_bot
# 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.