Supported tools, collection, and launch requirements
Usage collection, launching an agent, and reading plan limits are separate capabilities. A tool can provide local usage without being launchable from tokenstat. Launching an agent uses that tool's own installation, authentication, and permissions.
Compatibility
The launcher checks the machine where the project lives. For a remote workspace, the agent must be installed on that host. Development and preview builds can add support before it reaches a stable download. Use the changelog to check your version.
| Tool | Usage source | Agent launch profile | Plan-limit readings |
|---|---|---|---|
| Claude Code | Local session JSONL. Older rollup recovery | Yes | Vendor connection |
| Codex | Local rollout records | Yes | Local recorded windows |
| Grok Build | Local unified log and session records | Yes | Vendor connection |
| OpenCode | Local SQLite | Yes | Supported vendor connection |
| Cline | Local session metrics | Yes | Not provided |
| OpenClaw | Local session records | Yes | Not provided |
| Muse | Local event logs | Yes | Not provided |
| Devin CLI | Local SQLite | Yes | Not provided |
| Kimi Code | Local wire logs | Yes | Not provided |
| Qwen Code | Local usage ledger | Yes | Not provided |
| Pi | Local session JSONL | Yes | Not provided |
| DeepSeek Harness | Compressed session records | Yes | Not provided |
| Hermes Agent | Local SQLite rollups | Yes | Not provided |
| Kilo Code | Local SQLite, separate from OpenCode | Yes | Not provided |
| Copilot CLI | Local session usage records | Yes | Not provided |
| Zed | Local thread database | No dedicated profile | Not provided |
| Antigravity | CLI records locally. IDE fetch through the running app | CLI profile | App/vendor connection |
| Cursor | Vendor API fetch | Cursor Agent / CLI profiles | Vendor connection |
A listed plan-limit connection still needs usable credentials or local records. A missing reading does not mean the plan has no limit. Other agents may run as shell commands, but that does not add a dedicated parser or launch profile.
Find usage on this machine
tokenstat scan
tokenstat doctor
Collection reads supported locations on the current machine. The archive contains the records that are available there. It cannot infer a complete history from a missing source. See no sessions found for custom paths and troubleshooting.
Cursor usage requires a connection
Cursor's supported usage path is its API, rather than local token records. Configure authentication, fetch usage, and scan:
tokenstat auth cursor
tokenstat fetch
tokenstat scan
Authentication can use an available app credential or one you provide. Cursor and Antigravity setup explains the connection and its limits.
Antigravity has separate usage and quota readings
CLI records can be collected locally. IDE session collection needs the Antigravity app running so tokenstat can read its local service and cache usage for scanning. Plan quota is a separate status reading. It is not converted into token events.
Older Claude Code usage
Claude Code normally removes transcripts after 30 days. Its coarser rollup can recover some daily/model totals after the original records are gone. Recovered usage has less detail and is kept separate from transcript events in the archive.
Where detailed archive records already cover a day and model, tokenstat avoids adding the recovered total again. tokenstat doctor reports reconciliation and coverage.
Read the result
Token totals and estimated value depend on the fields each tool records. Confidence labels describe how repeated scans identify the same event. See what tokenstat counts and confidence labels.
More in Tools and sources