AI Pulse · Under the Radar

A tiny open-source tool called agentacct just picked up 539 GitHub stars in a week with no funding, no launch post, and no marketing. It does one thing: it reads the logs from Claude Code, Codex, and OpenCode, then shows you what each agent run actually cost. Tools called, files touched, tests run, tokens burned. Local dashboard, no login, no cloud.

The speed matters because it tells you something real. Developers are already running agents in their workflows, and the first question they ask is not “can I build more?” It’s “what did that just cost me?” Most agent frameworks give you a final bill or a vague token count. They do not show you the breakdown. Which tool ate the budget? Which file scan was redundant? Which test run could have been skipped? agentacct answers those questions by parsing the logs you already have, and people are adopting it fast because the alternative is flying blind.

This is the observability wedge. Agents are moving from demos to daily use, and the tooling that wins early adoption is not the flashiest model or the cleverest prompt. It’s the stuff that makes the black box a little less black. Cost visibility, task replay, failure traces. Boring infrastructure that lets you trust the thing enough to keep using it. If you’re building agents into your business, this is the kind of instrumentation you need baked in from the start, not bolted on later when the bill surprises you. It’s also the kind of thing we build into an AI command centre so you can see what’s running, what it’s costing, and where to tighten up before it scales.

The project is live on GitHub if you want to see what a no-frills observability layer looks like. The takeaway is not the tool itself. It’s that the market is already asking for this, and asking loudly.