Hey Product Hunt 👋

I kept handing Claude Code a task, walking away, and coming back to a session that had burned far more time and money than the job was worth. The agent has no sense of "this is enough."

Redline fixes that. You set a budget inside Claude Code:

/redline 10m $5

You get a live burn-down in your statusline, and the agent paces itself to land inside it.

Three things I cared about:

Never over. The budget is a hard ceiling.
Under is success. It lands around 90% with the task done.
Never abrupt. It always finishes with a usable result, no mid-sentence kill.

How it works: it's built entirely on Claude Cod so there's no wrapper, no daemon, and zerodependencies beyond the Node that ships with Claude Code. It emits the same token-budget signal the model is already trained to pace on, and translates time, dollars, and plan-clock deadline can drive the agent too. At 90% it locks new tool calls, so the agent spends its last slice of budget writing the answer instead of starting fresh work.

One budget covers everything the session spawns, including fanned-out subagents.

Free, MIT, open source: github.com/sankalpgunturi/redline

Would love feedback, especially on which budget you reach for first: time, dollars, or tokens.