Skip to content

Commands

On Windows, if sessioncost is not found, type python -m sessioncost instead (for example python -m sessioncost list).

Type these to your agent. In Claude Code with the plugin they may show as /sessioncost:sessioncost.

Type What it does
/sessioncost Builds the report of this chat and gives you its link.
/sessioncost clean Deletes the reports of older chats and keeps the latest.
/sessioncost update Updates every copy of SessionCost on this computer. Restart your agent afterwards.
/sessioncost version Shows which SessionCost version this agent uses, and how many other copies are on this computer.
/sessioncost auto off Stops rebuilding the report after each answer. /sessioncost auto on starts it again.
/livesession Claude Code: turns the guard on for every chat, this one included, until /livesession off.
/livesession strict The same, stricter: outside content stays untrusted for the whole chat.
/livesession off Turns the guard off for every chat. Only your own message does this: the agent cannot.

Run from the project folder whose chats you want to see. sessioncost below is typed python -m sessioncost if you installed with pip, or npx sessioncost with npm: see Which command to type.

Command What it does
sessioncost The report of the newest chat of this folder’s project, from any agent.
sessioncost list Your recent chats here: when, which agent, the chat id, cost, turns, the first message. --limit 20 for more.
sessioncost 3f2a9c1e The report of one chat: the first characters of its id from sessioncost list, or the path of a session file.
sessioncost setup Adds /sessioncost to every agent it finds. Restart the agents.
sessioncost clean Deletes the reports of older chats and keeps the latest.
sessioncost update Updates every copy. --dry-run shows the steps without running them.
sessioncost version This copy’s version, and how many other copies are on this computer. --all lists every copy (terminal, each agent’s plugin, each /sessioncost from setup) with its version and names the ones that differ.
sessioncost auto off Stops rebuilding each chat’s report after each answer. sessioncost auto on starts it again.

After each answer, SessionCost rebuilds that chat’s report on your computer, at most once a minute. Press F5 on the open report to see the latest. It uses no AI model, sends nothing, and adds no text the agent reads; the first time in a chat, you see one line saying where the report is. It is on by default and does not turn the guard on.

It works in Claude Code with the plugin, and in Codex, Cursor, Antigravity and Copilot CLI once you ran sessioncost guard setup <agent> for them (that is what gives SessionCost a signal at the end of each answer).

Add any of these after sessioncost or after a chat id. They combine, for example sessioncost --no-open --subscription.

Option What it does
--project FOLDER The newest chat of another project folder.
--no-open Writes the report but does not open the browser.
--brief Prints only the report’s path.
--subscription On a monthly plan (Claude Pro or Max, ChatGPT), money is labelled what it would cost at API prices.
--cursor-usage usage.csv Cursor only: your usage export, for exact tokens. It is also picked up from Downloads by itself.
--prices my-prices.json Your own prices laid over the built-in ones.
--json Every number as JSON. It is long: send it to a file, sessioncost --json > out.json.
--sarif FILE The risky steps (and the guard’s decisions) as SARIF 2.1.0, for security tools.
--version The installed version.
--help Every command and option.

The guard’s terminal commands are on Guard settings.