Two kinds of limit stop a chat:
- Your own ceilings, set in Settings > Chat > Spending: a dollar amount per chat, a dollar amount per project, a dollar amount per autopilot item, and a share of the model’s context.
- Your subscription’s usage limits, set by the agent’s vendor, such as Claude’s five hour and weekly windows. Tori shows how close each account is (see Accounts and usage) and can continue a chat once a limit resets.
Set a spending ceiling
- Open Settings (the gear in the title bar) and go to Chat, then the Spending group.
- Fill in any of the four fields. Leave a field blank for no limit, which is the default.
- Move Warn at to how full a ceiling may get before Tori warns you.
| Setting | Limits |
|---|---|
| Stop this chat after | Dollars one chat may spend |
| Stop this project after | Dollars across every chat in one project. Two chats on the same repository spend one budget |
| Stop an autopilot item after | Dollars across every session the autopilot runs for one item. See What work costs |
| Stop at context | Percent of the model’s context window one chat may fill |
| Warn at | How full a ceiling gets before Tori warns, from 50% to 100%. 80% by default. At 100% there is no early warning |
How a ceiling stops a chat
Ceilings are checked between turns, never in the middle of one. The turn that crosses a ceiling is always allowed to finish, so you never get half an edit. After that the chat starts no new turn:
- The transcript says This chat hit this chat’s spend limit of $5.00 ($5.12 used) and will not start another turn. Raise the limit in Settings to carry on., or the same for the project or context limit.
- A message you send is kept rather than refused. It goes into the queue, and the transcript says This message is waiting, with which limit was reached.
- Raise or clear the limit in Settings and the queued message goes out on its own. Nothing has to be typed again.
Before that, once a ceiling passes Warn at, the transcript warns once: This chat has used $4.10 of this chat’s spend limit ($5.00). It will stop when the limit is reached.
The context ceiling is the only one that recovers by itself: when the agent compacts the conversation and the context shrinks below the ceiling, the chat can carry on.
Nothing is ever written into the agent’s conversation to announce a ceiling. The agent does not know about your limits; Tori just stops starting turns.
Which agents ceilings work on
The dollar ceilings need the agent to report what each turn cost. Claude does. Codex, OpenCode and the other agents report how full the context is but not what it cost, so the ceilings are not offered for them rather than shown and never triggered. A chat’s Session panel lists spending ceilings under its capabilities when they apply.
When a usage limit stops a chat
When Claude stops because your subscription’s limit was reached, every chat on that account shows a banner above the transcript, for example Your rolling 5-hour limit has been reached. Resets 2:00 PM. A chat that has not run a turn yet shows it too, so you do not start work on an account that cannot answer.
Resume at reset
The banner can have Tori continue the chat for you once the limit resets:
- Click Resume at reset on the banner to do it for this chat. The banner adds Tori will continue this chat when it resets. Click Cancel to undo that.
- To make it the rule for every Claude chat, turn on Resume at reset in Settings > Chat. It is off by default.
Shortly after the limit resets, Tori sends the chat one message, Your usage limit has reset. Continue where you left off., shown in the transcript as a Tori row, Tori continued after the usage limit reset. The agent picks up where it stopped.
Some details:
- It is sent once per stop, and only to the chat that was stopped, not to every chat on the account.
- Your spending ceilings still apply. A chat over a ceiling is not continued, and says so.
- If a turn is already running when the limit resets, for example because you carried on by hand, nothing is sent.
- Messages you queued while the chat was stopped wait behind the resume: if the resumed turn finishes, they follow as usual.
- Tori has to be running when the limit resets. A relaunch forgets the plan.
- The stop and the continue both show in the Inbox’s Since you looked, one row per account.
- Only Claude chats can do this, and not chats the Autopilot is driving.
Too many chats at once
Every open chat is a running agent process with its own spend, even one in a background tab nobody is reading. Warn above, in the same Spending group, sets how many live chats are fine at once (4 by default). Past that, every chat shows 5 chats are running at once, past your limit of 4. Each one is a live agent process with its own token spend. Close one, or raise the limit in Settings. It is a warning, not a refusal; set it to 0 to turn it off.
The Autopilot keeps its own workers one under this number, so it fits too.
Two chats in one worktree
When a second chat starts in a worktree where another is already running, the new chat says Another chat is already running in this worktree. with a Got it button. Both chats edit the same files. Tori records which chat wrote what, so reverts and rewinds stay per chat, and a file both wrote is marked as shared rather than given to one of them. To keep their work apart, give each chat its own worktree.