Which agents
SessionCost reads each agent’s own log on your computer. How exact the numbers are depends on what that log records.
| Agent | Tokens |
|---|---|
| Claude Code | Exact, from its log. |
| Codex | Exact, from its log. |
| GitHub Copilot CLI | Exact, from its log. Money is API-equivalent: Copilot bills premium requests, not tokens. |
| GitHub Copilot Chat (VS Code) | Estimated from text size: the log has few token counts and no cache, so the cost is shown as a range. |
| Antigravity | Exact, from its log. |
| Cursor | Estimated from text size. Exact with your usage export: sessioncost --cursor-usage file.csv. |
The guard, by agent
Section titled “The guard, by agent”| Agent | How it gets the guard |
|---|---|
| Claude Code | With the plugin; turn it on with /livesession (every chat, until /livesession off). |
| Codex | sessioncost guard setup codex, then sessioncost guard on. Codex cannot ask, so a step that would ask is blocked and the agent asks you in the chat. |
| Cursor | sessioncost guard setup cursor, then sessioncost guard on. |
| Antigravity | sessioncost guard setup antigravity, then sessioncost guard on. |
| GitHub Copilot CLI and Copilot Chat (VS Code) | sessioncost guard setup copilot, then sessioncost guard on, then restart Copilot or VS Code. Copilot’s chat in VS Code runs the same hooks, so one setup covers both. It adds one file, ~/.copilot/hooks/sessioncost-guard.json; delete it to take the guard out. Copilot lets a step run if a check takes longer than 10 seconds, and in Autopilot mode a step that would ask is blocked instead. |
What each agent’s hooks let the guard see, side by side: What the guard checks.