--- title: Which agent is about to hit its limit? What each session used, as Claude Code and Codex counted it description: Right now shows the tokens, model and cost of each agent's session, and how much of each rate limit is left, read from the agents' own files. url: https://hamloy.com/blog/what-each-turn-used --- [Blog](https://hamloy.com/blog) # Which agent is about to hit its limit? What each session used, as Claude Code and Codex counted it Right now shows the tokens, model and cost of each agent's session, and how much of each rate limit is left, read from the agents' own files. 15 October 2026 · Rahmakyan Run four agents through an afternoon and two questions come up that no single terminal answers: which of them is about to run into its limit, and what did the afternoon use? Each agent knows its own numbers. You'd have to visit each window and ask. Hamloy's Right now page now answers both, beside each session: its tokens, its model, its cost where the agent gives one, and a strip with how much of each rate limit is used and when it starts over. ## Nobody asks the agent for it On a scratch project, Pebble sync, Claude Code was asked one thing: "Run npm test and tell me in one line how many tests pass." When it answered, its own result line said 4 tokens in, 193 out, 36,726 read from the cache and 5,741 written to it. Right now said the same sum beside the session: ``` HERE NOW codex weekly limit 2% used, starts over in 5 days Plus plan claude idle active just now laptop · ~/code/app · on main · not pushed yet 43k tokens · claude-haiku-5-5 codex idle active just now laptop · ~/code/app · on main · not pushed yet 1.4M tokens · gpt-5.6-luna ``` The Codex lines came from a real Codex session's own file: 1,434,707 tokens in, 1,350,912 of them from the cache, on the Plus plan. Its 5-hour window had already started over, so the strip showed only the weekly limit, and when it starts over. ## Hamloy counts nothing itself Every number on that page is one the agent's own program wrote down. When a turn ends, the [hamloy program](https://hamloy.com/docs/the-program) on your computer reads Claude Code's transcript or Codex's session file and sends what it says: the tokens, the model, and where the vendor gives them, the cost and the limits. During a long turn it sends them once a minute. Hamloy doesn't estimate tokens or price them with a table of its own. A number it would make up would be wrong in the ways that matter: cached reads, a model switched halfway, a plan's own limits. So the page shows what was said, and nothing where nothing was said. ## What each agent gives The two write down different things: - **Codex** gives tokens, the model, and its rate limits with the plan's name. It gives no cost. - **Claude Code** gives tokens and the model in its transcript. Its cost and limits go only to its status line, so they come once Hamloy's line is there: `hamloy statusline on`, with `--below` to keep your own line above it. Its limits come on a Pro or Max plan, and its cost is what the session would cost at list prices. A limit that is 80% used or more is drawn red. A limit is yours with the vendor across every project, so the strip takes the newest word from any of your sessions in the past day. ## Kept to yourself, if you like `hamloy config usage off` keeps what your sessions use on your computer, and the page still shows who works where. On a shared project everyone sees the same Right now, so they see what your sessions report. ## Why When you look after several agents, how much each has used is part of what each is doing. You decide which one to give the long task to by which one has room left, and you shouldn't have to visit every window to find out. See [Tokens, cost and rate limits](https://hamloy.com/docs/usage) for the details.