Hey everyone I'm launching TokenDam here tomorrow and wanted to share the story a day early.
I kept staring at agent token bills that felt too high but couldn't see why. Turns out the waste is boringly consistent: the same 2-3k-token system prompt resent uncached on every call, a whole scraped page dumped into context, 20+ MCP tools the model never touches, history that never gets compacted.
Your agents burn tokens — resending uncached prompts, dragging dead tools, replaying full history. TokenDam is a linter for LLM spend: paste a trace and see the line-item waste, each with a $/month figure and the exact fix (runs in your browser — your trace never leaves the tab). Put it in CI and the GitHub App audits every PR, fails the build on a cost regression, and opens a prompt-cache fix PR. Think ESLint, but for your token bill. Free CLI + web; Pro watches every PR.
