Hey Product Hunt 👋 I built Token Watcher because I kept running into the same problem while using agentic coding tools like Codex and Claude Code: token usage is powerful, but often invisible. When an agent is working well, it feels magical. But when it starts retrying, pulling in too much context, or looping through the same task, it can quietly burn through a lot of tokens before you notice. I wanted a lightweight way to see what was happening in my coding sessions without turning it into a full observability platform. Token Watcher helps developers: monitor token usage across coding-agent workflows spot unusually large prompts catch runaway loops earlier understand which sessions are getting expensive keep agentic coding more transparent and controllable The project is open source and intentionally simple. It is built for developers who are already using Codex, Claude Code, or similar AI coding CLIs and want better visibility into how their prompts and sessions are behaving. I’d love feedback from other builders using agentic coding tools: What token/cost visibility do you wish these tools had by default? Would this fit into your workflow? What should I add next? Thanks for checking it out.

Tokenwatch - Codex/Claude Code local token watcher

Replies