Get it for macOS

See what every agent thread costs in dollars while it runs, per prompt and across all your threads. A chip beside the context circle shows the thread's total and what the last prompt added; click it for where the money went, a chart of every model call and the total with child threads. The Burn tab in the side panel lists the cost of each prompt, and a Burn page in bb's navigation charts spend across threads by day, project, model and harness. Budgets notify you at 80% and 100% of a thread's budget, and stop a run only if you turn that on. Works with pi, OpenCode, Codex, Claude Code and any harness that reports usage to bb; where a harness records no cost, Burn estimates it from list prices and marks it with ~.


Overview

View source

What you get

  • A cost chip under the prompt box, beside the context circle. It shows the thread's total and, while the thread works, what the last prompt added, for example $4.27 +$0.30. Click it for a card with where the money went (input, cache write, cache read and output), a bar for every model call, the models used, and the total with the thread's child threads, such as minions. From the card you can set a budget, open the Burn tab or copy a summary. On a phone, a flame in the thread header opens the same card while the prompt box is collapsed.
  • A Burn tab in the side panel that lists every prompt with its cost, newest or costliest first, and every child thread with its model, calls and cost.
  • A Burn page in bb's navigation with spend across all threads for today, the last 7 days or the last 30 days, or any earlier day, 7 days or 30 days, for every project or one. It ranks by cost, tokens or calls, and shows spend per day by model, the token mix against what it cost, call times, the busiest hours, the costliest threads, a table of models, and spend by project and by harness.
  • A Cost entry in every thread menu, and a red flame in the sidebar on threads over their budget.

Budgets

Set a budget for one thread from the chip, the thread menu or the Burn tab, or set one budget for every thread in the plugin settings. Burn sends a notification at 80% and at 100% of a thread's budget, and once a day when all threads together pass the daily budget you set. It never stops a thread unless you turn on Stop a thread at its budget, which is off by default. Another optional notification flags a prompt that cost at least three times the thread's median and at least $0.50.

Where the numbers come from

Burn counts every thread, whatever harness runs it. For pi threads it reads the session logs that bb's pi bridge keeps on each machine, where pi already records the cost of every call. For OpenCode threads it reads opencode's own database on the same machine, where every assistant message records its tokens and cost, including the calls of subagents that the task tool starts in child sessions. For Codex, Claude Code and other harnesses it reads the token usage that bb records after each call or turn, and keeps its own copy, because bb deletes older usage events. When a harness reports no cost, Burn prices its tokens at list price and marks the amount as an estimate, such as ~$1.23. A model with no known price shows tokens instead of dollars.

The pi extension (optional)

The same repository ships a pi extension, installed separately with pi install git:github.com/kalindudc/burn. It adds a cost pill to pi's status line and a /burn overlay in the terminal, with or without bb. In a bb thread it also stops a run at its budget from inside pi, which is quicker than stopping it through bb. The bb plugin works without it.

Requirements and limits

  • bb 0.46 or newer. No account and no outside service: Burn reads only what bb and pi already record.
  • Dollar figures are list prices unless you set a discount in Burn's settings, such as a contract rate.
  • Claude Code threads update once per turn, do not count Claude Code's subagents, and are never stopped at their budget, though budget alerts still fire.
  • Call times are recorded for pi and OpenCode threads; failed calls only when the harness records that the call failed.
  • A harness that sends bb no usage, and whose own records Burn does not read, shows nothing.

Documentation and source: https://github.com/kalindudc/burn

More from Kalindu De Costa

FlareLet an agent tell you what it found while it is still working, not only in the message that ends its turn. Agents get a `flare` tool and send a finding, a decision, a progress or status update, a blocker or a summary; the newest one shows in a banner above the message box, and the thread's Flares tab keeps them all, drawn the way bb draws a reply, so an agent can answer with a chart, a diagram, a table or a snippet instead of describing it. Blockers, and flares the agent marks urgent, send a push notification, so alerts reach you while the work carries on. A child thread's flares appear in its parent's tab, and a child's blocker reaches the parent's agent as a note without starting a turn. The tool returns at once, so the agent never waits, and `bb flare send` works from a terminal or another agent.NewForagePut web research in every agent's hands and keep what it found in a per-thread Research tab you can read, quote and cite. web_search, web_fetch, web_map, web_extract, web_crawl and web_diff run as bb's own tools for every provider, with routing profiles, domain and date filters, bounded limits and every partial failure in words. Replies carry citation cards that quote the exact kept lines a claim rests on, and Check in sources matches any message against the pages stored for that thread. A zero-turn CLI runs the same operations; private networks are blocked by default and keyless DuckDuckGo and Exa search need no account.NewLoomHand a goal to agents and keep the decisions. A planner turns it into a spec with tickets, acceptance criteria, checks and a merge plan, and a separate agent reviews the spec before you approve it. Workers then build each ticket in its own git worktree, Loom runs your checks, and reviewers on another model approve the work or send findings back. Approved work merges into an integration branch you land as one pull request, or opens as one pull request per ticket that Loom follows through CI. A full-screen Loom page shows what needs you, the board, every attempt and review, the changes before they land, insights on how the project went and a timeline you can rewind.NewSession LogCheck what an agent actually ran, changed and decided, in one live log beside the chat. Messages, thinking, commands with their output and exit codes, tool calls, file diffs and subagent steps appear in timeline order, with a timestamp on every row and a duration on every command. The log follows new output as it arrives, opens failed commands by default, and filters to messages, thinking, tools or errors. Search covers the whole session, and single keys jump between rows, replies and errors. Open it with Mod+Shift+L or from the thread header; it reads bb's timeline and keeps no copy of the thread.New

More in Token Usage & Limits 34

Understand or control token, context-window, and provider-quota use.

View all